CN107169565B - Spinning quality prediction method for improving BP neural network based on firework algorithm - Google Patents

Spinning quality prediction method for improving BP neural network based on firework algorithm Download PDF

Info

Publication number
CN107169565B
CN107169565B CN201710288559.2A CN201710288559A CN107169565B CN 107169565 B CN107169565 B CN 107169565B CN 201710288559 A CN201710288559 A CN 201710288559A CN 107169565 B CN107169565 B CN 107169565B
Authority
CN
China
Prior art keywords
firework
network
neural network
spinning
fwa
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.)
Expired - Fee Related
Application number
CN201710288559.2A
Other languages
Chinese (zh)
Other versions
CN107169565A (en
Inventor
邵景峰
马创涛
马晓红
杨小渝
王蕊超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Polytechnic University
Original Assignee
Xian Polytechnic University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xian Polytechnic University filed Critical Xian Polytechnic University
Priority to CN201710288559.2A priority Critical patent/CN107169565B/en
Publication of CN107169565A publication Critical patent/CN107169565A/en
Application granted granted Critical
Publication of CN107169565B publication Critical patent/CN107169565B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Feedback Control In General (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a spinning quality prediction method for improving a BP (back propagation) neural network based on a firework algorithm, which is characterized in that the firework algorithm is introduced into the BP neural network, the network weight and the threshold of the BP neural network model are optimized by utilizing the optimization mechanism of the firework algorithm, input and output indexes are selected, a spinning quality prediction model based on FWA-BP is constructed, the spinning quality prediction model based on FWA-BP established in the step 2 is learned and trained by utilizing a data set subjected to standardized processing, and finally the prediction of spinning quality is completed. The method solves the problem that the spinning quality is difficult to accurately predict due to numerous factors influencing the yarn quality in a spinning system and mutual coupling, can effectively establish the function mapping relation between the fiber index and the yarn quality, realizes the prediction of the yarn quality in spinning production, and is beneficial to improving the quality management level of a spinning workshop.

Description

Spinning quality prediction method for improving BP neural network based on firework algorithm
Technical Field
The invention belongs to the technical field of spinning quality prediction and control, and relates to a spinning quality prediction method for improving a BP neural network based on a firework algorithm.
Background
The spinning system is in a complex environment with a plurality of factors such as high temperature, high humidity, high electromagnetism and the like staggered with each other, the factors have a coupling effect relationship with each other, and the spinning production and processing process is complex in flow and raw materials are frequently subjected to a physicochemical modification process, so that the quality prediction in the textile production process is more challenging compared with the quality prediction of the traditional pure mechanical processing. Particularly, the fiber attribute indexes are increased in a geometric shape and reach more than 300 at present, the yarn quality factors influencing the factors in the spinning system are numerous and have coupling relations with each other, and in addition, the fiber attributes and the yarn quality characteristic values form a nonlinear correlation relation, so that the prediction result of the spinning quality prediction model is established by utilizing a neural network under the training of small sample data, and the actual requirements of the production management of a spinning workshop are difficult to meet.
With the improvement of the informatization degree of spinning production, a large amount of yarn quality data such as raw materials, processes, equipment and the like are accumulated in the textile production process, so that the establishment of a spinning quality prediction model based on a neural network under the environment of large sample data becomes possible. However, in an environment of a large amount of training sample data, as the number of input neurons and the amount of training sample data in the neural network prediction model are greatly increased, the problems that the neural network model is low in convergence speed and easily falls into local optimization are further highlighted, and the accuracy of spinning quality prediction is greatly restricted.
Disclosure of Invention
The invention aims to provide a spinning quality prediction method for improving a BP (back propagation) neural network based on a firework algorithm, which solves the problems of low prediction precision and high iteration times in the training process of the existing neural network model.
The invention discloses a spinning quality prediction method for improving a BP neural network based on a firework algorithm, which is implemented according to the following steps:
step 1, optimizing the network weight and the threshold of a BP neural network model by utilizing an optimization mechanism of a firework algorithm, and establishing a FWA-BP neural network model based on firework algorithm optimization;
step 2, selecting input and output indexes on the basis of the FWA-BP neural network model in the step 1, and constructing a spinning quality prediction model based on the FWA-BP;
step 3, learning and training the spinning quality prediction model based on FWA-BP established in the step 2 by using a data set subjected to standardization processing, and finally completing the prediction of the spinning quality;
the specific steps of constructing a spinning quality prediction model based on FWA-BP in the step 2 are as follows:
step 2.1, selecting input and output indexes: selecting raw materials and process data related to yarn quality in the spinning production and processing process as input variables, selecting the CV value of the yarn as an output index, and inputting and outputting the whole FWA-BP-based yarn quality prediction model as follows:
the input quantity is: x1 is the sliver content, x2 is the roving twist coefficient, x3 is the moisture regain, x4 is the fiber diameter, x5 is the fiber length, x6 is the diameter dispersion coefficient, x7 is the fiber mass irregularity, x8 is the fiber drafting multiple, x9 is the spun yarn wire loop number, and x10 is the roller rotation speed; the output quantity is as follows: y is the yarn CV value;
2.2, establishing a data set of a model according to the input and output data obtained in the step 2.1, and carrying out standardization processing on data in the data set by using a Min-Max method;
step 2.3, determining a strategy of a network structure, determining the number of input, output and hidden layer layers according to the input and output indexes selected in the step 2.1, wherein the number m of nodes of the input layer of the FWA-BP spinning quality prediction model is 10, the number n of nodes of the output layer is 1, and the number of hidden layer neurons is determined according to the following formula
Figure GDA0002427824850000031
Calculating to obtain s-6;
and 2.4, selecting an activation function, wherein the input layer adopts a tansig activation function, the output layer adopts a purelin activation function, and the trainlm function is selected as a training function of the network model.
The present invention is also characterized in that,
the optimization mechanism of the firework algorithm in the step 1 optimizes the network weight and the threshold of the BP neural network model, and the method comprises the following specific steps:
step 1.1, key parameter coding, namely selecting a coding strategy of real number vectors to code key parameters in the model, and recording a vector X ═ X1,x2,Λ,xD]Representing a group of parameters to be optimized, wherein each dimensional vector of the parameters consists of network weight and threshold, and the dimension of the firework population is as follows: d ═ nIW(1,1)+nb(1,1)+nIW(2,1)+nb(2,1)Wherein, note nIW(1,1)N is the number of weights between the hidden layer and the output layerb(1,1)For the number of hidden layer neuron thresholds, nIW(2,1)To be hiddenNumber of weights between the containing layer and the output layer, nb(2,1)The number of neuron thresholds of the output layer;
step 1.2, initializing the weight coefficient and the threshold, and utilizing the firework individual x in the firework algorithm on the basis of the step 1.1ikThe position of (a) represents a neuron in the neural network, and a weight coefficient between the ith neuron and the jth neuron in the l layer of the network in the kth iteration process in the neural network
Figure GDA0002427824850000032
And a threshold value thetaiInitial coding into vector X ═ X1,x2,Λ,xD]And using a random initialization strategy to initialize the vector X in the interval [ -1,1 [ -1 [ ]]Inner, then have weight coefficient wij～U[-1,1]，
Wherein i refers to the weight of the ith neuron node in the network, j refers to the weight of the jth neuron node, l refers to the number of network layers where the current weight is located, and k refers to the current iteration number;
step 1.3, calculating the error of the firework individual, introducing a fitness function, and calculating a square error SSE by using a formula (1) and a formula (2), wherein the formula (1) and the formula (2) are as follows:
Figure GDA0002427824850000041
where t is the expected output of the network, p is the number of layers of the network, s is the number of network output units, and y is the network output value, which is specifically expressed as follows:
Figure GDA0002427824850000042
wherein x isjBeing an input to the network, wijAs a weight of a network node, θiIs the threshold value of the ith neuron in the network and thetai＝-wi(n+1)；
Step 1.4, each firework individual x obtained by calculation in step 1.3iOn the basis of errors, f is introducedi(x) Function is madeCalculating the fitness value of each individual xi of the vector X in the step 1.2 through the fitness function as a fitness function, wherein the fitness function is shown as a formula (3) as follows,
Figure GDA0002427824850000043
step 1.5, optimizing the firework population, and on the basis of the step 1.4, aiming at each firework individual xiCarrying out explosion, displacement and variation operations, wherein the explosion variation operation and the Gaussian variation mapping rule are formula (4) to formula (6),
h＝Ai×rand(1,-1) (4)
exik＝xik+h (5)
mxik＝xik×e (6)
wherein A isiIs the explosion radius of the ith firework, h is the position offset, xikThe kth dimension, ex, representing the ith Firework in the populationikFor the spark after explosion of the ith fireworks, mxikIs xikGaussian variation sparks after Gaussian variation, and Gaussian distribution of e to N (1, 1);
step 1.6, selecting the next generation of firework population, and regarding the firework individuals x subjected to the operations of explosion, displacement and variation in the step 1.5iCalculating each firework individual x by using the formula in the step 1.4iSelecting the optimal firework individual to form the next generation firework population by using the selection strategies of the formula (7) and the formula (8), wherein the specific selection strategy is as follows:
selecting min (f (x) with minimum fitness valuei) ) individual xkDirectly using a wheel roulette mode for one firework population individual and the rest N-1 firework individuals to select the candidate individual xiThe probability of its selection is as follows:
Figure GDA0002427824850000051
wherein R (x)i) Representing individual fireworks xiDistance from other individualsSum of the following formulae;
Figure GDA0002427824850000052
step 1.7, judging a termination condition, and calculating the fitness value f (x) of the firework individuals in the firework population according to a formula (3) and a formula (8)i) And the European distance R (x) between the firework unitsi) And judging whether the maximum iteration times reached in the termination condition are met or not, and if so, calculating to obtain the minimum fitness value min (f (x) of the firework individuals in the current firework populationi) And the maximum distance max (R (x)) between firework individuals in the firework populationi) And taking the current firework population as the optimal firework population XbestOtherwise, continuing to execute the step 1.3;
step 1.8, optimizing network weight and threshold value, and obtaining optimal firework population X in step 1.7bestAnd (3) initializing the weight and the threshold value in the corresponding neural network in the vector X in the step 1.2.
In the step 3, the specific steps of learning and predicting the spinning quality prediction model based on FWA-BP established in the step 2 by utilizing the data set subjected to standardization processing are as follows:
3.1, selecting a strategy of training a data set, namely selecting 80% of the data set as the training data set and the rest 20% of the data set as the test data set by using the data set subjected to standardization processing in the step 2.2;
step 3.2, setting key parameters in the firework algorithm, wherein the size N of the firework population is 70, the firework explosion radius regulating constant D is 5, the firework explosion spark number regulating constant m is 40, the upper limit value lm of the firework explosion spark number is 0.8, the lower limit value bm of the firework explosion spark number is 0.04, the gaussian variation spark number g is 5, and the maximum iteration number T is 100, wherein the dimension D of the variable is 85, the total number of the neuron weight and the threshold value in the network model is taken on the basis of the step 1.1, and specifically, the total number of the neuron weight and the threshold value in the network model is calculated on the basis of the step 2.3 through the following formula
D＝m×s+s×n+s+n＝10×7+7×1+7+1＝85
Wherein, m, s and n are the numbers of input layer neurons, hidden layer neurons and output layer neurons of the network respectively;
3.3, on the basis of the setting of the firework algorithm parameters in the step 3.2, training a spinning quality prediction model based on the FWA-BP by using the training data set selected in the step 3.1, wherein the related parameters are set in the network training process, the learning rate is 0.01, the momentum factor is 0.9, the maximum iteration number is 20000, and the training minimum error is 0.05;
and 3.4, obtaining a spinning quality prediction model based on FWA-BP through training in the steps 3.1-3.3, and performing test statistical analysis and experimental simulation on the prediction effect of the model by using the test data set selected in the step 3.1.
Compared with the prior art, the invention has the following effects: the invention improves the accuracy of spinning quality prediction and reduces the iteration times of the network. The invention mainly introduces the firework algorithm into the neural network model, and optimizes the weight and the threshold of the neural network model by utilizing the mechanism of simultaneous diffusion of multiple points in the firework explosion process, thereby reducing the iteration times of the prediction model and improving the accuracy of model prediction.
Drawings
FIG. 1 is a flow chart of a spinning quality prediction method of the invention based on a firework algorithm to improve a BP neural network;
FIG. 2 is a simulation result diagram of the predicted value and the actual value of the spinning quality in the embodiment of the spinning quality prediction method based on the firework algorithm improved BP neural network;
FIG. 3 is a graph of simulation results comparing spinning quality prediction results of an embodiment of the method for predicting spinning quality of a BP neural network based on fireworks algorithm improvement with spinning quality prediction results of other BP neural networks, GA-BP neural networks and PSO-BP neural networks;
FIG. 4 is a graph of correlation analysis of mapping relationships between input variables and output variables established based on a BP neural network, obtained by training with the same parameters in an embodiment of the present invention;
FIG. 5 is a graph of correlation analysis of mapping relationships between input variables and output variables established based on GA-BP neural network, obtained by the same parameter training in an embodiment of the present invention;
FIG. 6 is a graph of correlation analysis of mapping relationships between input variables and output variables established based on a PSO-BP neural network, obtained by the same parameter training in an embodiment of the present invention;
fig. 7 is a correlation analysis diagram of the mapping relationship between the input variable and the output variable established based on the FWA-BP neural network, which is proposed in the embodiment of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
In the embodiment of the invention, as shown in fig. 1, the spinning quality prediction method based on the firework algorithm improved BP neural network is implemented according to the following steps:
step 1, optimizing the network weight and the threshold of a BP neural network model by utilizing an optimization mechanism of a firework algorithm, and establishing a FWA-BP neural network model based on firework algorithm optimization, wherein the optimization mechanism of the firework algorithm optimizes the network weight and the threshold of the BP neural network model by the following specific steps:
step 1.1, key parameter coding, namely selecting a coding strategy of real number vectors to code key parameters in the model, and recording a vector X ═ X1,x2,Λ,xD]Representing a group of parameters to be optimized, wherein each dimensional vector of the parameters consists of network weight and threshold, and the dimension of the firework population is as follows: d ═ nIW(1,1)+nb(1,1)+nIW(2,1)+nb(2,1)Wherein, note nIW(1,1)N is the number of weights between the hidden layer and the output layerb(1,1)For the number of hidden layer neuron thresholds, nIW(2,1)N is the number of weights between the hidden layer and the output layerb(2,1)The number of neuron thresholds of the output layer;
step 1.2, initializing the weight coefficient and the threshold, and utilizing the firework individual x in the firework algorithm on the basis of the step 1.1ikRepresents a neuron in a neural network, the first in the neural networkWeight coefficient between ith neuron and jth neuron in i layer of k times of iterative process network
Figure GDA0002427824850000081
And a threshold value thetaiInitial coding into vector X ═ X1,x2,Λ,xD]And using a random initialization strategy to initialize the vector X in the interval [ -1,1 [ -1 [ ]]Inner, then have weight coefficient wij～U[-1,1]，
Wherein i refers to the weight of the ith neuron node in the network, j refers to the weight of the jth neuron node, l refers to the number of network layers where the current weight is located, and k refers to the current iteration number;
step 1.3, calculating the error of the firework individual, introducing a fitness function, and calculating a square error SSE by using a formula (1) and a formula (2), wherein the formula (1) and the formula (2) are as follows:
Figure GDA0002427824850000082
where t is the expected output of the network, p is the number of layers of the network, s is the number of network output units, and y is the network output value, which is specifically expressed as follows:
Figure GDA0002427824850000083
wherein x isjBeing an input to the network, wijAs a weight of a network node, θiIs the threshold value of the ith neuron in the network and thetai＝-wi(n+1)；
Step 1.4, each firework individual x obtained by calculation in step 1.3iOn the basis of errors, f is introducedi(x) The function is used as a fitness function, and each firework individual X of the vector X in the step 1.2 is calculated through the fitness functioniThe fitness function is as shown in equation (3) below,
Figure GDA0002427824850000091
step 1.5, optimizing the firework population, and on the basis of the step 1.4, aiming at each firework individual xiCarrying out explosion, displacement and variation operations, wherein the explosion variation operation and the Gaussian variation mapping rule are formula (4) to formula (6),
h＝Ai×rand(1,-1) (4)
exik＝xik+h (5)
mxik＝xik×e (6)
wherein A isiIs the explosion radius of the ith firework, h is the position offset, xikThe kth dimension, ex, representing the ith Firework in the populationikFor the spark after explosion of the ith fireworks, mxikIs xikGaussian variation sparks after Gaussian variation, and Gaussian distribution of e to N (1, 1);
step 1.6, selecting the next generation of firework population, and regarding the firework individuals x subjected to the operations of explosion, displacement and variation in the step 1.5iCalculating each firework individual x by using the formula in the step 1.4iSelecting the optimal firework individual to form the next generation firework population by using the selection strategies of the formula (7) and the formula (8), wherein the specific selection strategy is as follows:
selecting min (f (x) with minimum fitness valuei) ) individual xkDirectly using a wheel roulette mode for one firework population individual and the rest N-1 firework individuals to select the candidate individual xiThe probability of its selection is as follows:
Figure GDA0002427824850000092
wherein R (x)i) Representing individual fireworks xiThe sum of the distances to other individuals, specifically the following formula;
Figure GDA0002427824850000093
step 1.7, judging the termination condition, and calculating the fireworks according to the formula (3) and the formula (8)Fitness value f (x) of firework individual in populationi) And the European distance R (x) between the firework unitsi) And judging whether the maximum iteration times reached in the termination condition are met or not, and if so, calculating to obtain the minimum fitness value min (f (x) of the firework individuals in the current firework populationi) And the maximum distance max (R (x)) between firework individuals in the firework populationi) And taking the current firework population as the optimal firework population XbestOtherwise, continuing to execute the step 1.3;
step 1.8, optimizing network weight and threshold value, and obtaining optimal firework population X in step 1.7bestAnd (3) initializing the weight and the threshold value in the corresponding neural network in the vector X in the step 1.2.
Step 2, on the basis of the FWA-BP neural network model in the step 1, selecting input and output indexes, constructing a spinning quality prediction model based on the FWA-BP, and specifically constructing the spinning quality prediction model based on the FWA-BP comprises the following steps:
step 2.1, selecting input and output indexes:
considering that the yarn quality is affected by raw materials, process parameters and equipment parameters under the mutual coupling action of multiple factors in the spinning production and processing process, 10 indexes shown in the following table are selected from the aspects of raw material performance, process parameters, equipment parameters and the like as the input of a spinning quality prediction model, and the yarn CV value is selected as the output index of the spinning quality prediction model.
Figure GDA0002427824850000101
2.2, establishing a data set of a model according to the input and output data obtained in the step 2.1, and carrying out standardization processing on data in the data set by using a Min-Max method;
the normalization of the data is accomplished by the following equation:
Figure GDA0002427824850000102
wherein max (x) is the maximum value in the training data set, min (x) is the maximum value in the training data set, and after the data is subjected to standardization processing, the training data is mapped to the interval [0,1], so that comprehensive evaluation and comparison are facilitated.
Step 2.3, determining a strategy of a network structure, determining the number of input, output and hidden layer layers according to the input and output indexes selected in the step 2.1, wherein the number m of nodes of the input layer of the FWA-BP spinning quality prediction model is 10, the number n of nodes of the output layer is 1, and the number of hidden layer neurons is determined according to the following formula
Figure GDA0002427824850000111
Wherein, m is 10 and n is 1, which are the number of input layer nodes and the number of output layer nodes of the network respectively, and s is 6;
and 2.4, selecting an activation function, wherein the input layer adopts a tansig activation function, the output layer adopts a purelin activation function, and the trainlm function is selected as a training function of the network model.
Step 3, learning and training the spinning quality prediction model based on FWA-BP established in the step 2 by using the data set subjected to standardization processing, and finally completing the prediction of the spinning quality, wherein the specific steps are as follows:
and 3.1, training a selection strategy of a data set, taking cotton spinning quality data of a certain company, and carrying out experimental verification on the effectiveness of the FWA-BP spinning quality prediction model algorithm. In the training process of the algorithm model, the first 80% of data of the data set is taken as a training data set for training the FWA-BP model, and the last 20% of data of the data set is taken as a test set for testing the model predictive performance;
step 3.2, setting key parameters in the firework algorithm according to the weight value of the network to be optimized
Figure GDA0002427824850000112
And a threshold value thetaiThe method is characterized in that the specific optimization target is combined with the experimental results in related documents, the key parameters in the firework algorithm are set to be 70 for the size N of the firework population, 5 for the firework explosion radius regulating constant d, and the firework explosion is carried outThe method is characterized in that a regulating constant m of the number of the explosion sparks is 40, an upper limit value lm of the number of the fireworks explosion sparks is 0.8, a lower limit value bm of the number of the fireworks explosion sparks is 0.04, a number g of the Gaussian variation sparks is 5, the maximum iteration number T is 100, a dimension D of a variable is 85, the total number of neuron weights and threshold values in a network model is taken on the basis of a step 1.1, and specifically, the total number of the neuron weights and the threshold values in the network model is calculated on the basis of a step 2.3 through the following formula
D＝m×s+s×n+s+n＝10×7+7×1+7+1＝85
Wherein, m, s and n are the numbers of input layer neurons, hidden layer neurons and output layer neurons of the network respectively;
3.3, on the basis of the setting of the firework algorithm parameters in the step 3.2, training a spinning quality prediction model based on the FWA-BP by using the training data set selected in the step 3.1, wherein the related parameters are set in the network training process, the learning rate is 0.01, the momentum factor is 0.9, the maximum iteration number is 20000, and the training minimum error is 0.05;
and 3.4, obtaining a spinning quality prediction model based on FWA-BP through training in the steps 3.1-3.3, and performing test statistical analysis and experimental simulation on the prediction effect of the model by using the test data set selected in the step 3.1.
In addition, the same training set data subjected to the standardization processing in the step 2 are used for training spinning quality prediction models such as a traditional BP neural network, GA-BP and PSO-BP, and error rates and iteration times of prediction results of different algorithms are calculated.
In order to reduce accidental factors in the experimental process, the same data is used for training and testing the same algorithm model for 10 times, the average values of the predicted error and the iteration times of 10 times are respectively taken as the predicted error value and the convergence speed of the algorithm to be evaluated, the training result of the FWA-BP based yarn quality prediction model is shown in Table 1, and the following table 1 shows that: compared with a particle swarm optimization neural network (PSO-BP), the spinning quality prediction method based on the FWA-BP neural network has the advantages that the error rate of spinning quality characteristic value fluctuation prediction is reduced by 49.52%, the prediction precision is 97.88%, and the iteration frequency of the algorithm is reduced by 31.11%.
TABLE 1 training results of FWA-BP based yarn quality prediction model
Figure GDA0002427824850000131
As shown in FIG. 2, the simulation result diagram of the predicted value and the actual value of the spinning quality in the embodiment of the invention shows that the spinning quality prediction model based on the GA-BP neural network provided in the embodiment of the invention can better realize the prediction of the spinning quality;
compared with the spinning quality prediction results of other BP neural networks, GA-BP neural networks and PSO-BP neural networks, the embodiment of the invention has a simulation result diagram, as shown in FIG. 3, it can be seen that the prediction result of the FWA-BP based neural network model for the spinning quality is closer to an actual value;
in the embodiment of the present invention, a correlation analysis graph of the mapping relationship between the input variable and the output variable established based on the BP neural network is obtained by the same parameter training, as shown in fig. 4, and the correlation coefficient R is 0.85176;
in the embodiment of the present invention, a correlation analysis graph of the mapping relationship between the input variable and the output variable established based on the GA-BP neural network, which is obtained by the same parameter training, is shown in fig. 5, and the correlation coefficient R thereof is 0.91472;
in the embodiment of the invention, a correlation analysis chart of the mapping relation between the input variable and the output variable established based on the PSO-BP neural network is obtained through the same parameter training, as shown in FIG. 6, the correlation coefficient R is 0.92182;
the correlation analysis diagram of the mapping relationship between the input variable and the output variable established based on the FWA-BP neural network proposed in the embodiment of the present invention is shown in fig. 7, and the correlation coefficient R is 0.9479.
The method introduces a firework algorithm into optimization of weight and threshold of a neural network, provides a prediction model based on the FWA-BP network, and experiments and simulation results show that the method provided by the invention has lower prediction error rate and fewer iteration times, can effectively solve the problems of low prediction precision and high iteration times of the traditional neural network prediction model, and provides a new method for quickly and effectively solving the prediction problem under large sample data.

Claims (3)

1. The spinning quality prediction method for improving the BP neural network based on the firework algorithm is characterized by comprising the following steps:
step 1, optimizing the network weight and the threshold of a BP neural network model by utilizing an optimization mechanism of a firework algorithm, and establishing a FWA-BP neural network model based on firework algorithm optimization;
step 2, selecting input and output indexes on the basis of the FWA-BP neural network model in the step 1, and constructing a spinning quality prediction model based on the FWA-BP;
step 3, learning and training the spinning quality prediction model based on FWA-BP established in the step 2 by using a data set subjected to standardization processing, and finally completing the prediction of the spinning quality;
the specific steps of constructing a spinning quality prediction model based on FWA-BP in the step 2 are as follows:
step 2.1, selecting input and output indexes: selecting raw materials and process data related to yarn quality in the spinning production and processing process as input variables, selecting the CV value of the yarn as an output index, and inputting and outputting the whole FWA-BP-based yarn quality prediction model as follows:
the input quantity is: x1 is the sliver content, x2 is the roving twist coefficient, x3 is the moisture regain, x4 is the fiber diameter, x5 is the fiber length, x6 is the diameter dispersion coefficient, x7 is the fiber mass irregularity, x8 is the fiber drafting multiple, x9 is the spun yarn wire loop number, and x10 is the roller rotation speed; the output quantity is as follows: y is the yarn CV value;
2.2, establishing a data set of a model according to the input and output data obtained in the step 2.1, and carrying out standardization processing on data in the data set by using a Min-Max method;
step 2.3, determining a strategy of a network structure, determining the number of input, output and hidden layer layers according to the input and output indexes selected in the step 2.1, wherein the number m of nodes of the input layer of the FWA-BP spinning quality prediction model is 10, the number n of nodes of the output layer is 1, and the number of hidden layer neurons is determined according to the following formula
Figure FDA0002427824840000011
Calculating to obtain s-6;
and 2.4, selecting an activation function, wherein the input layer adopts a tansig activation function, the output layer adopts a purelin activation function, and the trainlm function is selected as a training function of the network model.
2. The spinning quality prediction method for improving the BP neural network based on the firework algorithm as claimed in claim 1, wherein the optimization mechanism of the firework algorithm in the step 1 for optimizing the network weight and the threshold of the BP neural network model comprises the following specific steps:
step 1.1, key parameter coding, namely selecting a coding strategy of real number vectors to code key parameters in the model, and recording a vector X ═ X1,x2,Λ,xD]Representing a group of parameters to be optimized, wherein each dimensional vector of the parameters consists of network weight and threshold, and the dimension of the firework population is as follows: d ═ nIW(1,1)+nb(1,1)+nIW(2,1)+nb(2,1)Wherein, note nIW(1,1)N is the number of weights between the hidden layer and the output layerb(1,1)For the number of hidden layer neuron thresholds, nIW(2,1)N is the number of weights between the hidden layer and the output layerb(2,1)The number of neuron thresholds of the output layer;
step 1.2, initializing the weight coefficient and the threshold, and utilizing the firework individual x in the firework algorithm on the basis of the step 1.1ikThe position of (a) represents a neuron in the neural network, and a weight coefficient between the ith neuron and the jth neuron in the l layer of the network in the kth iteration process in the neural network
Figure FDA0002427824840000021
And a threshold value thetaiInitial coding into vector X ═ X1,x2,Λ,xD]And using a random initialization strategy to initialize the vector X in the interval [ -1,1 [ -1 [ ]]Inner, then have weight coefficient wij～U[-1,1]，
Wherein i refers to the weight of the ith neuron node in the network, j refers to the weight of the jth neuron node, l refers to the number of network layers where the current weight is located, and k refers to the current iteration number;
step 1.3, calculating the error of the firework individual, introducing a fitness function, and calculating a square error SSE by using a formula (1) and a formula (2), wherein the formula (1) and the formula (2) are as follows:
Figure FDA0002427824840000031
where t is the expected output of the network, p is the number of layers of the network, s is the number of network output units, and y is the network output value, which is specifically expressed as follows:
Figure FDA0002427824840000032
wherein x isjBeing an input to the network, wijAs a weight of a network node, θiIs the threshold value of the ith neuron in the network and thetai＝-wi(n+1)；
Step 1.4, each firework individual x obtained by calculation in step 1.3iOn the basis of errors, f is introducedi(x) The function is used as a fitness function, and each firework individual X of the vector X in the step 1.2 is calculated through the fitness functioniThe fitness function is as shown in equation (3) below,
Figure FDA0002427824840000033
step 1.5, optimizing the firework population, and on the basis of the step 1.4, aiming at each firework individual xiPerforming explosion, displacement and mutation operations, wherein the explosion mutation operation and the Gaussian mutation mapping rule are formula (4) -publicThe compound of the formula (6),
h＝Ai×rand(1,-1) (4)
exik＝xik+h (5)
mxik＝xik×e (6)
wherein A isiIs the explosion radius of the ith firework, h is the position offset, xikThe kth dimension, ex, representing the ith Firework in the populationikFor the spark after explosion of the ith fireworks, mxikIs xikGaussian variation sparks after Gaussian variation, and Gaussian distribution of e to N (1, 1);
step 1.6, selecting the next generation of firework population, and regarding the firework individuals x subjected to the operations of explosion, displacement and variation in the step 1.5iCalculating each firework individual x by using the formula in the step 1.4iSelecting the optimal firework individual to form the next generation firework population by using the selection strategies of the formula (7) and the formula (8), wherein the specific selection strategy is as follows:
selecting min (f (x) with minimum fitness valuei) ) individual xkDirectly using a wheel roulette mode for one firework population individual and the rest N-1 firework individuals to select the candidate individual xiThe probability of its selection is as follows:
Figure FDA0002427824840000041
wherein R (x)i) Representing individual fireworks xiThe sum of the distances to other individuals, specifically the following formula;
Figure FDA0002427824840000042
step 1.7, judging a termination condition, and calculating the fitness value f (x) of the firework individuals in the firework population according to a formula (3) and a formula (8)i) And the European distance R (x) between the firework unitsi) And judging whether the maximum iteration times reached in the termination condition are met or not, and if so, calculating to obtain the current firework varietyMinimum fitness value min (f (x) of individual fireworks in the groupi) And the maximum distance max (R (x)) between firework individuals in the firework populationi) And taking the current firework population as the optimal firework population XbestOtherwise, continuing to execute the step 1.3;
step 1.8, optimizing network weight and threshold value, and obtaining optimal firework population X in step 1.7bestAnd (3) initializing the weight and the threshold value in the corresponding neural network in the vector X in the step 1.2.
3. The spinning quality prediction method based on the firework algorithm improved BP neural network as claimed in claim 1, wherein the step 3 of learning and predicting the spinning quality prediction model based on FWA-BP established in step 2 by using the standardized data set comprises the following specific steps:
3.1, selecting a strategy of training a data set, namely selecting 80% of the data set as the training data set and the rest 20% of the data set as the test data set by using the data set subjected to standardization processing in the step 2.2;
step 3.2, setting key parameters in the firework algorithm, wherein the size N of the firework population is 70, the firework explosion radius regulating constant D is 5, the firework explosion spark number regulating constant m is 40, the upper limit value lm of the firework explosion spark number is 0.8, the lower limit value bm of the firework explosion spark number is 0.04, the gaussian variation spark number g is 5, and the maximum iteration number T is 100, wherein the dimension D of the variable is 85, the total number of the neuron weight and the threshold value in the network model is taken on the basis of the step 1.1, and specifically, the total number of the neuron weight and the threshold value in the network model is calculated on the basis of the step 2.3 through the following formula
D＝m×s+s×n+s+n＝10×7+7×1+7+1＝85
Wherein, m, s and n are the numbers of input layer neurons, hidden layer neurons and output layer neurons of the network respectively;
3.3, on the basis of the setting of the firework algorithm parameters in the step 3.2, training a spinning quality prediction model based on the FWA-BP by using the training data set selected in the step 3.1, wherein the related parameters are set in the network training process, the learning rate is 0.01, the momentum factor is 0.9, the maximum iteration number is 20000, and the training minimum error is 0.05;
and 3.4, obtaining a spinning quality prediction model based on FWA-BP through training in the steps 3.1-3.3, and performing test statistical analysis and experimental simulation on the prediction effect of the model by using the test data set selected in the step 3.1.
CN201710288559.2A 2017-04-27 2017-04-27 Spinning quality prediction method for improving BP neural network based on firework algorithm Expired - Fee Related CN107169565B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710288559.2A CN107169565B (en) 2017-04-27 2017-04-27 Spinning quality prediction method for improving BP neural network based on firework algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710288559.2A CN107169565B (en) 2017-04-27 2017-04-27 Spinning quality prediction method for improving BP neural network based on firework algorithm

Publications (2)

Publication Number Publication Date
CN107169565A CN107169565A (en) 2017-09-15
CN107169565B true CN107169565B (en) 2020-06-19

Family

ID=59813156

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710288559.2A Expired - Fee Related CN107169565B (en) 2017-04-27 2017-04-27 Spinning quality prediction method for improving BP neural network based on firework algorithm

Country Status (1)

Country Link
CN (1) CN107169565B (en)

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107395433B (en) * 2017-08-18 2020-06-09 中南大学 Wireless sensor node deployment method based on firework algorithm
CN107994570B (en) * 2017-12-04 2022-03-25 华北电力大学（保定） State estimation method and system based on neural network
CN108241298A (en) * 2018-01-09 2018-07-03 南京航空航天大学 A kind of aerogenerator method for diagnosing faults based on FWA-RNN models
CN108197696A (en) * 2018-01-31 2018-06-22 湖北工业大学 A kind of network navy account recognition methods and system
CN109636014B (en) * 2018-11-28 2020-05-05 郑州轻工业学院 Cotton blending method based on finished yarn quality prediction
CN109542103B (en) * 2018-12-25 2019-12-20 北京理工大学 Robot welding path planning method based on firework particle swarm algorithm
CN110032069B (en) * 2019-04-02 2020-09-15 东华大学 Polyester fiber spinning process segmented parameter configuration method based on error compensation
CN113112101B (en) * 2020-01-10 2023-07-25 北京百度网讯科技有限公司 Method and device for monitoring production process, electronic equipment and storage medium
CN111277141A (en) * 2020-02-29 2020-06-12 武汉理工大学 Bidirectional DC/DC converter optimization control method
CN111460738B (en) * 2020-04-16 2023-06-16 中南大学 RNN-ARX modeling method and RNN-ARX model of magnetic suspension system
CN112529684A (en) * 2020-11-27 2021-03-19 百维金科(上海)信息科技有限公司 Customer credit assessment method and system based on FWA _ DBN
CN112733720A (en) * 2021-01-12 2021-04-30 上海理工大学 Face recognition method based on firework algorithm improved convolutional neural network
CN112765902B (en) * 2021-02-09 2024-02-20 嘉兴学院 Soft measurement modeling method for COD concentration in rural domestic sewage treatment process based on TentFWA-GD RBF neural network
CN113408963A (en) * 2021-07-28 2021-09-17 上海致景信息科技有限公司 Textile yarn quality rating method and device, storage medium and processor
CN114065631B (en) * 2021-11-18 2024-06-28 福州大学 Energy consumption prediction method and system for plate laser cutting
CN113919601A (en) * 2021-12-09 2022-01-11 山东捷瑞数字科技股份有限公司 Resin process prediction method and device based on product performance and process data model
CN115099490A (en) * 2022-06-24 2022-09-23 无锡物联网创新中心有限公司 Yarn quality prediction method and related device
CN115221791A (en) * 2022-07-27 2022-10-21 浙江大学 Supercritical boiler wall temperature online prediction method and system
CN114994289B (en) * 2022-08-01 2022-10-28 江苏卓鹏智能机电有限公司 Spinning product quality detection method and system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105652952A (en) * 2016-04-18 2016-06-08 中国矿业大学 Maximum power point tracking method for photovoltaic power generation system based on fireworks algorithm
CN205688082U (en) * 2016-04-26 2016-11-16 苏州精卫智能科技有限公司 Yarn qualities predictor
CN106503661A (en) * 2016-10-25 2017-03-15 陕西师范大学 Face gender identification method based on fireworks depth belief network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105652952A (en) * 2016-04-18 2016-06-08 中国矿业大学 Maximum power point tracking method for photovoltaic power generation system based on fireworks algorithm
CN205688082U (en) * 2016-04-26 2016-11-16 苏州精卫智能科技有限公司 Yarn qualities predictor
CN106503661A (en) * 2016-10-25 2017-03-15 陕西师范大学 Face gender identification method based on fireworks depth belief network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
《Artificial Neural Network Training using Fireworks》;Ram Kinkar Dutta等;《internationaljournal of computer applications》;20160430;1-4 *
《基于ＢＰ神经网络组合预测的纺纱产量预测研究》;刘有时等;《江苏纺织》;20121120;５５－６０ *
《烟花算法优化极限学习机的研究及应用》;苌康群;《中国秀硕士论文辑信息科技辑》;20170215;I140-108页 *

Also Published As

Publication number Publication date
CN107169565A (en) 2017-09-15

Similar Documents

Publication Publication Date Title
CN107169565B (en) Spinning quality prediction method for improving BP neural network based on firework algorithm
CN111738512B (en) Short-term power load prediction method based on CNN-IPSO-GRU hybrid model
CN110544011B (en) Intelligent system combat effectiveness evaluation and optimization method
CN107395430B (en) Cloud platform dynamic risk access control method
CN112085254B (en) Prediction method and model based on multi-fractal cooperative measurement gating circulation unit
Huang et al. Stochastic ranking for offline data-driven evolutionary optimization using radial basis function networks with multiple kernels
CN115277354B (en) Fault detection method for command control network management system
CN111079926B (en) Equipment fault diagnosis method with self-adaptive learning rate based on deep learning
CN106650022A (en) Method for predicting fault of complex electronic device
CN113707228B (en) Wet flue gas desulfurization optimization method based on LightGBM algorithm
CN109784692A (en) A kind of fast and safely constraint economic load dispatching method based on deep learning
CN114202252A (en) Multi-target optimization method for spun yarn process parameters
CN113887694A (en) Click rate estimation model based on characteristic representation under attention mechanism
CN114004153A (en) Penetration depth prediction method based on multi-source data fusion
CN106453294A (en) Security situation prediction method based on niche technology with fuzzy elimination mechanism
CN115481727A (en) Intention recognition neural network generation and optimization method based on evolutionary computation
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
CN110032069A (en) A kind of polyester fiber spinning process segmentation parameter configuration method based on error compensation
CN109697531A (en) A kind of logistics park-hinterland Forecast of Logistics Demand method
CN109063837A (en) Genetic algorithm information flow network property analysis method based on complex network structures entropy
CN117507365A (en) Artificial intelligent control method based on dynamic self-adaptive parameter adjustment of materials and environment
CN114943866B (en) Image classification method based on evolutionary neural network structure search
CN116956160A (en) Data classification prediction method based on self-adaptive tree species algorithm
Zhao et al. Research on glass relics based on machine learning
CN114048837A (en) Deep neural network model reinforcement method based on distributed brain-like map

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
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200619

Termination date: 20210427