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

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

Improved BP Neural Network Based on Fireworks Algorithm for Spinning Quality Prediction Method

technical field

The invention belongs to the technical field of spinning quality prediction and control, and relates to a spinning quality prediction method based on a firework algorithm improved BP neural network.

Background technique

The spinning system is in a complex environment where various factors such as high temperature, high humidity and high electromagnetics are intertwined, and there is a coupling relationship between each factor. The modification process makes quality prediction during textile production more challenging compared to traditional purely mechanical quality prediction. In particular, the fiber attribute index has increased geometrically, and has reached more than 300 at present, and there are many factors affecting yarn quality in the spinning system and there is a coupling relationship between them. In addition, there is a relationship between fiber attributes and yarn quality characteristics. The nonlinear correlation makes it difficult to use the neural network to establish the prediction results of the spinning quality prediction model under the training of small sample data, and it is difficult to meet the actual requirements of the production management of the spinning workshop.

With the improvement of the informatization of spinning production, a large amount of yarn quality data such as raw materials, processes, and equipment has been accumulated in the textile production process, which makes it possible to establish a neural network-based spinning quality prediction model in the environment of large sample data. However, in the environment of a large number of training sample data, with the large increase in the number of input neurons and the amount of training sample data in the neural network prediction model, the problem of slow convergence of the neural network model and easy to fall into local optimum is further highlighted. To a large extent, it restricts the accuracy of spinning quality prediction.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a spinning quality prediction method based on the fireworks algorithm improved BP neural network, which solves the problems of low prediction accuracy and high iteration times in the training process of the existing neural network model.

The present invention improves the spinning quality prediction method of BP neural network based on fireworks algorithm, and is specifically implemented according to the following steps:

Step 1, using the optimization mechanism of the fireworks algorithm to optimize the network weight and threshold of the BP neural network model, and establish a FWA-BP neural network model optimized based on the fireworks algorithm;

Step 2, on the basis of the FWA-BP neural network model in step 1, select input and output indicators to construct a spinning quality prediction model based on FWA-BP;

Step 3, use the standardized data set to learn and train the spinning quality prediction model based on FWA-BP established in step 2, and finally complete the prediction of spinning quality;

The specific steps of constructing the spinning quality prediction model based on FWA-BP in step 2 are as follows:

Step 2.1, the selection of input and output indicators: select the raw materials and process data related to the yarn quality in the spinning production and processing process as input variables, and select the CV value of the yarn as the output indicator, then the whole yarn quality based on FWA-BP The input and output of the prediction model are:

The input quantities are: x1 = sliver trash content, x2 = roving twist coefficient, x3 = moisture regain, x4 = fiber diameter, x5 = fiber length, x6 = diameter dispersion coefficient, x7 = fiber quality unevenness, x8 = fiber Drafting ratio, x9=spinning traveler number, x10=roller rotation speed; output: Y=yarn CV value;

Step 2.2, establish a data set of the model according to the input and output data obtained in step 2.1, and use the Min-Max method to standardize the data in the data set;

Step 2.3, determine the strategy of the network structure, according to the input and output indicators selected in step 2.1, determine the number of layers of the input, output and hidden layers, the number of nodes in the input layer of the FWA-BP spinning quality prediction model m=10, The number of output layer nodes is n=1, and the number of hidden layer neurons is determined by the following formula

Calculated to get s=6;

Step 2.4, the selection of the activation function, the input layer adopts the tansig activation function, the output layer adopts the purelin activation function, and the trainlm function is selected as the training function of the network model.

The present invention is also characterized in that,

The specific steps for optimizing the network weights and thresholds of the BP neural network model by the optimization mechanism of the fireworks algorithm in step 1 are:

Step 1.1, key parameter coding, select the coding strategy of the real number vector to encode the key parameters in the model, and note that the vector X=[x ₁ , x ₂ , Λ, x _D ] represents a set of parameters to be optimized, each dimension of which is The vector consists of network weights and thresholds, and the dimension of the firework population is: D=n _IW(1,1) +n _b(1,1) +n _IW(2,1) +n _b(2,1) , where , denoted n _IW(1,1) as the number of weights between the hidden layer and the output layer, n _b(1,1) as the number of hidden layer neuron thresholds, n _IW(2,1) as The number of weights between the hidden layer and the output layer, the number of n _b(2,1) output layer neuron thresholds;

和阈值θ_i初始化编码成向量X＝[x₁,x₂,Λ,x_D]，并利用随机初始化的策略把向量X初始在区间[-1,1]内，则有权重系数w_ij～U[-1,1]，Step 1.2: Initialize the weight coefficient and threshold. On the basis of step 1.1, the position of the firework individual x _ik in the firework algorithm is used to represent the neuron in the neural network, and the k-th iteration process in the neural network is used. The weight coefficient between the neuron and the jth neuron

and the threshold θ _i are initialized and encoded into a vector X=[x ₁ ,x ₂ ,Λ,x _D ], and the random initialization strategy is used to initialize the vector X in the interval [-1,1], then there are weight coefficients w _ij ~ U[-1,1],

Among them, i refers to the weight of the ith neuron node in the network, j refers to the weight of the jth neuron node, l represents the number of network layers where the current weight is located, and k represents the current the number of iterations;

Step 1.3, calculate the error of the individual fireworks, introduce the fitness function and use the formula (1) and formula (2) to calculate the square error SSE, formula (1) and formula (2) are as follows:

Among them, t is the expected output of the network, p is the number of layers of the network, s is the number of output units of the network, and y is the output value of the network, which is as follows:

Among them, x _j is the input of the network, w _ij is the weight of the network node, θ _i is the threshold of the ith neuron in the network and θ _i = _-wi (n+1);

Step 1.4, based on the error of each firework individual _xi calculated in step 1.3, introduce the f _i (x) function as the fitness function, and calculate the fitness of each firework individual xi in the vector X in step 1.2 through the fitness function. value, the fitness function is shown in formula (3) as follows,

Step 1.5, firework population optimization, on the basis of step 1.4, perform explosion, displacement and mutation operations for each firework individual _xi , wherein the explosion mutation operation and the Gaussian mutation mapping rule are formula (4) ~ formula (6) ,

h=A _i ×rand(1,-1) (4)

ex _ik = x _ik +h (5)

mx _ik = x _ik ×e (6)

Among them, A _i is the explosion radius of the i-th firework, h is the position offset, x _ik is the k-th dimension of the i-th firework in the population, ex _ik is the spark of the i-th firework after the explosion, and mx _ik is the The Gaussian mutation spark of x _ik after Gaussian mutation, the Gaussian distribution of e～N(1,1);

Step 1.6, select the next generation firework population, for the firework individual _xi after the explosion, displacement and mutation operations in step 1.5, use the formula in step 1.4 to calculate the fitness value of each firework individual _xi , and use formula (7 ) and the selection strategy of formula (8), select the optimal firework individual to form the next generation of fireworks population, and the specific selection strategy is:

Select the min(f( _xi )) individual x _k with the smallest fitness value as the following firework population, and the remaining N-1 firework individuals adopt the roulette method. For the candidate individual x _i , the probability of being selected is as follows: :

Among them, R( _xi ) represents the sum of the distances between the firework individual _xi and other individuals, and the specific formula is as follows;

Step 1.7, determine the termination condition, calculate the fitness value f(x _i ) of the individual fireworks in the firework population and the Euclidean distance R(x _i ) between the individual fireworks according to formula (3) and formula (8), and determine whether the termination condition is The maximum number of iterations reached in the firework population, if it is satisfied, the minimum fitness value min(f(x _i )) of the firework individuals in the current firework population and the maximum distance between the firework individuals in the firework population max(R(x _i ) ), and take the current fireworks population as the optimal fireworks population X _best , otherwise continue to step 1.3;

Step 1.8, optimize the network weights and thresholds, and use the optimal firework population X _best obtained in step 1.7 to initialize the weights and thresholds in the neural network corresponding to the vector X in step 1.2.

In step 3, the specific steps for learning and predicting the spinning quality prediction model based on FWA-BP established in step 2 using the standardized data set are as follows:

Step 3.1, the selection strategy of the training data set, using the standardized data set in step 2.2, select 80% of the data set as the training data set, and the remaining 20% of the data set as the test data set;

Step 3.2, the setting of key parameters in the fireworks algorithm, the size of the fireworks population N=70, the adjustment constant of the fireworks explosion radius d=5, the adjustment constant of the number of fireworks explosion sparks m=40, the upper bound value of the number of fireworks explosion sparks lm=0.8, The lower bound value of the number of fireworks explosion sparks bm=0.04, the number of Gaussian variation sparks g=5, the maximum number of iterations T=100, and the dimension of the variable D=85, which is based on step 1.1. The weight of neurons in the network model is obtained and the total number of thresholds, which are calculated by the following formula on the basis of step 2.3

D=m×s+s×n+s+n=10×7+7×1+7+1=85

Among them, m, s, n are the number of input layer neurons, hidden layer neurons and output layer neurons of the network respectively;

Step 3.3, on the basis of the fireworks algorithm parameter setting in step 3.2, use the training data set selected in step 3.1 to train the spinning quality prediction model based on FWA-BP, and the relevant parameters in the network training process are set as, The learning rate is 0.01, the momentum factor is 0.9, the maximum number of iterations is 20000, and the minimum training error is 0.05;

In step 3.4, the spinning quality prediction model based on FWA-BP is obtained through the training in steps 3.1 to 3.3, and the test data set selected in step 3.1 is used to perform test statistical analysis and experimental simulation on the prediction effect of the model.

Compared with the prior art, the present invention has the following effects: the present invention improves the accuracy of spinning quality prediction and reduces the number of iterations of the network. The invention mainly introduces the firework algorithm into the neural network model, and optimizes the weight and threshold of the neural network model by using the mechanism of simultaneous diffusion of multiple points in the firework explosion process, thereby reducing the number of iterations of the prediction model and improving the prediction accuracy of the model. Accuracy.

Description of drawings

Fig. 1 is the flow chart of the spinning quality prediction method that the present invention improves BP neural network based on fireworks algorithm;

Fig. 2 is the simulation result diagram of the spinning quality prediction value and the actual value of the embodiment in the spinning quality prediction method based on the firework algorithm improvement BP neural network of the present invention;

Fig. 3 is based on the firework algorithm improvement BP neural network in the spinning quality prediction method of the present invention and other BP neural network, GA-BP neural network and PSO-BP neural network spinning quality prediction result contrast simulation result diagram;

Fig. 4 is the correlation analysis diagram of the mapping relationship between the input variable and the output variable established based on the BP neural network obtained through the same parameter training in the embodiment of the present invention;

Fig. 5 is the correlation analysis diagram of the mapping relationship between the input variable and the output variable established based on the GA-BP neural network obtained through the same parameter training in the embodiment of the present invention;

Fig. 6 is the correlation analysis diagram of the mapping relationship between the input variable and the output variable established based on the PSO-BP neural network obtained through the same parameter training in the embodiment of the present invention;

FIG. 7 is a correlation analysis diagram of the mapping relationship between input variables and output variables established based on the FWA-BP neural network proposed in the embodiment of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

In the embodiment of the present invention, the flow chart of the method for predicting the spinning quality of the BP neural network based on the fireworks algorithm is improved.

Step 1: Use the optimization mechanism of the fireworks algorithm to optimize the network weights and thresholds of the BP neural network model, and establish a FWA-BP neural network model optimized based on the fireworks algorithm. The specific steps for optimizing network weights and thresholds are:

Step 1.1, key parameter coding, select the coding strategy of the real number vector to encode the key parameters in the model, and note that the vector X=[x ₁ , x ₂ , Λ, x _D ] represents a set of parameters to be optimized, each dimension of which is The vector consists of network weights and thresholds, and the dimension of the firework population is: D=n _IW(1,1) +n _b(1,1) +n _IW(2,1) +n _b(2,1) , where , denoted n _IW(1,1) as the number of weights between the hidden layer and the output layer, n _b(1,1) as the number of hidden layer neuron thresholds, n _IW(2,1) as The number of weights between the hidden layer and the output layer, the number of n _b(2,1) output layer neuron thresholds;

和阈值θ_i初始化编码成向量X＝[x₁,x₂,Λ,x_D]，并利用随机初始化的策略把向量X初始在区间[-1,1]内，则有权重系数w_ij～U[-1,1]，Step 1.2: Initialize the weight coefficient and threshold. On the basis of step 1.1, the position of the firework individual x _ik in the firework algorithm is used to represent the neuron in the neural network, and the k-th iteration process in the neural network is used. The weight coefficient between the neuron and the jth neuron

and the threshold θ _i are initialized and encoded into a vector X=[x ₁ ,x ₂ ,Λ,x _D ], and the random initialization strategy is used to initialize the vector X in the interval [-1,1], then there are weight coefficients w _ij ~ U[-1,1],

Among them, i refers to the weight of the ith neuron node in the network, j refers to the weight of the jth neuron node, l represents the number of network layers where the current weight is located, and k represents the current the number of iterations;

Step 1.3, calculate the error of the individual fireworks, introduce the fitness function and use the formula (1) and formula (2) to calculate the square error SSE, formula (1) and formula (2) are as follows:

Among them, t is the expected output of the network, p is the number of layers of the network, s is the number of output units of the network, and y is the output value of the network, which is as follows:

Among them, x _j is the input of the network, w _ij is the weight of the network node, θ _i is the threshold of the ith neuron in the network and θ _i = _-wi (n+1);

Step 1.4, on the basis of the error of each firework individual x _i calculated in step 1.3, introduce the f _i (x) function as a fitness function, and calculate the fitness of each firework individual x _i of the vector X in step 1.2 through the fitness function. degree value, the fitness function is shown in formula (3) as follows,

Step 1.5, firework population optimization, on the basis of step 1.4, perform explosion, displacement and mutation operations for each firework individual _xi , wherein the explosion mutation operation and the Gaussian mutation mapping rule are formula (4) ~ formula (6) ,

h=A _i ×rand(1,-1) (4)

ex _ik = x _ik +h (5)

mx _ik = x _ik ×e (6)

Among them, A _i is the explosion radius of the i-th firework, h is the position offset, x _ik is the k-th dimension of the i-th firework in the population, ex _ik is the spark of the i-th firework after the explosion, and mx _ik is the The Gaussian mutation spark of x _ik after Gaussian mutation, the Gaussian distribution of e～N(1,1);

Step 1.6, select the next generation firework population, for the firework individual _xi after the explosion, displacement and mutation operations in step 1.5, use the formula in step 1.4 to calculate the fitness value of each firework individual _xi , and use formula (7 ) and the selection strategy of formula (8), select the optimal firework individual to form the next generation of fireworks population, and the specific selection strategy is:

Select the min(f( _xi )) individual x _k with the smallest fitness value as the following firework population, and the remaining N-1 firework individuals adopt the roulette method. For the candidate individual x _i , the probability of being selected is as follows: :

Among them, R( _xi ) represents the sum of the distances between the firework individual _xi and other individuals, and the specific formula is as follows;

Step 1.7, determine the termination condition, calculate the fitness value f(x _i ) of the individual fireworks in the firework population and the Euclidean distance R(x _i ) between the individual fireworks according to formula (3) and formula (8), and determine whether the termination condition is The maximum number of iterations reached in the firework population, if it is satisfied, the minimum fitness value min(f(x _i )) of the firework individuals in the current firework population and the maximum distance between the firework individuals in the firework population max(R(x _i ) ), and take the current fireworks population as the optimal fireworks population X _best , otherwise continue to step 1.3;

Step 1.8, optimize the network weights and thresholds, and use the optimal firework population X _best obtained in step 1.7 to initialize the weights and thresholds in the neural network corresponding to the vector X in step 1.2.

Step 2, on the basis of the FWA-BP neural network model in step 1, select the input and output indicators, and construct a spinning quality prediction model based on FWA-BP. The specific steps for constructing a spinning quality prediction model based on FWA-BP are as follows: :

Step 2.1, the selection of input and output indicators:

Considering that the spinning production and processing process is under the interaction of multiple factors, the raw materials, process and equipment parameters will have an impact on the yarn quality. Therefore, from the aspects of raw material performance, process parameters, and equipment parameters, the 10 shown in the following table are selected. The index is used as the input of the spinning quality prediction model, and the yarn CV value is selected as the output index of the spinning quality prediction model.

Step 2.2, establish a data set of the model according to the input and output data obtained in step 2.1, and use the Min-Max method to standardize the data in the data set;

Data normalization is done by the following formula:

Among them, max(X) is the maximum value in the training data set, and min(X) is the maximum value in the training data set. After standardizing the data, the training data is mapped to the interval [0,1], which is convenient for comprehensive evaluation. Compared.

Step 2.3, determine the strategy of the network structure, according to the input and output indicators selected in step 2.1, determine the number of layers of the input, output and hidden layers, the number of nodes in the input layer of the FWA-BP spinning quality prediction model m=10, The number of output layer nodes is n=1, and the number of hidden layer neurons is determined by the following formula

Among them, m=10 and n=1 are the number of input layer nodes and the number of output layer nodes of the network, respectively, and s=6 is calculated;

Step 2.4, the selection of the activation function, the input layer adopts the tansig activation function, the output layer adopts the purelin activation function, and the trainlm function is selected as the training function of the network model.

Step 3, use the standardized data set to learn and train the spinning quality prediction model based on FWA-BP established in step 2, and finally complete the prediction of spinning quality. The specific steps are:

Step 3.1, the selection strategy of the training data set, take the cotton spinning quality data of a company, and verify the effectiveness of the FWA-BP spinning quality prediction model algorithm. In the training process of the algorithm model, the first 80% of the data set is taken as the training data set, which is used to train the FWA-BP model, and the last 20% of the data set is taken as the test set, which is used to predict the model. performance test;

和阈值θ_i的具体优化目标，并结合相关文献中的实验结果，对烟花算法中的关键参数设置为，烟花种群的大小N＝70，烟花爆炸半径调节常数d＝5，烟花爆炸火花数调节常数m＝40，烟花爆炸火花个数上界值lm＝0.8，烟花爆炸火花个数下界值bm＝0.04，高斯变异火花数g＝5,最大迭代次数T＝100，变量的维数D＝85，是在步骤1.1的基础之上取网络模型中神经元权重和阈值的总数，具体是在步骤2.3的基础之上通过如下公式计算Step 3.2, the setting of key parameters in the fireworks algorithm, according to the weight value of the network to be optimized

and the specific optimization objective of the threshold θ _i , combined with the experimental results in the relevant literature, the key parameters in the firework algorithm are set as: the size of the firework population N=70, the firework explosion radius adjustment constant d=5, the firework explosion spark number adjustment The constant m=40, the upper bound value of the number of fireworks explosion sparks lm=0.8, the lower bound value of the number of fireworks explosion sparks bm=0.04, the number of Gaussian variation sparks g=5, the maximum number of iterations T=100, the dimension of the variable D=85 , is to take the total number of neuron weights and thresholds in the network model on the basis of step 1.1, specifically calculated by the following formula on the basis of step 2.3

D=m×s+s×n+s+n=10×7+7×1+7+1=85

Among them, m, s, n are the number of input layer neurons, hidden layer neurons and output layer neurons of the network respectively;

Step 3.3, on the basis of the fireworks algorithm parameter setting in step 3.2, use the training data set selected in step 3.1 to train the spinning quality prediction model based on FWA-BP. The relevant parameters in the network training process are set as follows, The learning rate is 0.01, the momentum factor is 0.9, the maximum number of iterations is 20000, and the minimum training error is 0.05;

In step 3.4, the spinning quality prediction model based on FWA-BP is obtained through the training in steps 3.1 to 3.3, and the test data set selected in step 3.1 is used to perform test statistical analysis and experimental simulation on the prediction effect of the model.

In addition, use the same training set data that has been standardized in step 2 to train the traditional BP neural network, GA-BP, and PSO-BP and other spinning quality prediction models, and calculate the error rate and iteration of the prediction results of different algorithms frequency.

In order to reduce the chance factor in the experimental process, the same algorithm model is trained and tested 10 times with the same data, and the average value of the 10 prediction errors and iterations is taken as the prediction error value and convergence speed of the algorithm. Based on FWA The training results of the yarn quality prediction model of -BP are shown in Table 1. It can be seen from Table 1 that the spinning quality prediction method based on the FWA-BP neural network proposed in this specific implementation is better than the neural network optimized by particle swarm optimization. (PSO-BP), the error rate of eigenvalue fluctuation prediction of spinning quality decreased by 49.52%, the prediction accuracy reached 97.88%, and the number of iterations of the algorithm was reduced by 31.11%.

Table 1 Training results of the yarn quality prediction model based on FWA-BP

The simulation result diagram of the spinning quality prediction value and the actual value of the embodiment of the present invention is shown in Figure 2. From the figure, it can be seen that the spinning quality prediction model based on the GA-BP neural network proposed in the embodiment of the present invention is , which can better predict the spinning quality;

The simulation results of the comparison of the spinning quality prediction results of the embodiment of the present invention and other BP neural networks, GA-BP neural networks and PSO-BP neural networks are shown in Figure 3. It can be seen from the figure that the FWA-BP-based The prediction result of the neural network model for the spinning quality is closer to the actual value;

In the embodiment of the present invention, the correlation analysis diagram of the mapping relationship between the input variable and the output variable established based on the BP neural network obtained by training with the same parameters, as shown in FIG. 4 , the correlation coefficient R is 0.85176;

The correlation analysis diagram of the mapping relationship between the input variable and the output variable based on the GA-BP neural network established by training with the same parameters in the embodiment of the present invention, as shown in Figure 5, the correlation coefficient R is 0.91472;

In the embodiment of the present invention, the correlation analysis diagram of the mapping relationship between the input variable and the output variable established based on the PSO-BP neural network is obtained by training with the same parameters, as shown in FIG. 6 , the correlation coefficient R is 0.92182;

The correlation analysis diagram of the mapping relationship between input variables and output variables 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 invention introduces the fireworks algorithm into the optimization of the neural network weights and thresholds, and proposes a prediction model based on the FWA-BP network. The experimental and simulation results show that the method proposed by the invention has a lower prediction error rate and a higher prediction error rate. With a small number of iterations, it can effectively solve the problems of low prediction accuracy and high number of iterations in traditional neural network prediction models, and provide a new method for quickly and effectively solving prediction problems under large sample data.

Claims (3)

1. improve the spinning quality prediction method of BP neural network based on fireworks algorithm, it is characterized in that, specifically implement according to the following steps: Step 1, using the optimization mechanism of the fireworks algorithm to optimize the network weight and threshold of the BP neural network model, and establish a FWA-BP neural network model optimized based on the fireworks algorithm; Step 2, on the basis of the FWA-BP neural network model in step 1, select input and output indicators to construct a spinning quality prediction model based on FWA-BP; Step 3, use the standardized data set to learn and train the spinning quality prediction model based on FWA-BP established in step 2, and finally complete the prediction of spinning quality; The specific steps of constructing the FWA-BP-based spinning quality prediction model in the step 2 are: Step 2.1, the selection of input and output indicators: select the raw materials and process data related to the yarn quality in the spinning production and processing process as input variables, and select the CV value of the yarn as the output indicator, then the whole yarn quality based on FWA-BP The input and output of the prediction model are: The input quantities are: x1 = sliver trash content, x2 = roving twist coefficient, x3 = moisture regain, x4 = fiber diameter, x5 = fiber length, x6 = diameter dispersion coefficient, x7 = fiber quality unevenness, x8 = fiber Drafting ratio, x9=spinning traveler number, x10=roller rotation speed; output: Y=yarn CV value; Step 2.2, establish a data set of the model according to the input and output data obtained in step 2.1, and use the Min-Max method to standardize the data in the data set; Step 2.3, determine the strategy of the network structure, according to the input and output indicators selected in step 2.1, determine the number of layers of the input, output and hidden layers, the number of nodes in the input layer of the FWA-BP spinning quality prediction model m=10, The number of output layer nodes is n=1, and the number of hidden layer neurons is determined by the following formula

Calculated to get s=6; Step 2.4, the selection of the activation function, the input layer adopts the tansig activation function, the output layer adopts the purelin activation function, and the trainlm function is selected as the training function of the network model. 2. the spinning quality prediction method based on fireworks algorithm improvement BP neural network according to claim 1, is characterized in that, in described step 1, the optimization mechanism of fireworks algorithm optimizes the network weight and the threshold value of BP neural network model The specific steps are: Step 1.1, key parameter coding, select the coding strategy of the real number vector to encode the key parameters in the model, and note that the vector X=[x ₁ , x ₂ , Λ, x _D ] represents a set of parameters to be optimized, each dimension of which is The vector consists of network weights and thresholds, and the dimension of the firework population is: D=n _IW(1,1) +n _b(1,1) +n _IW(2,1) +n _b(2,1) , where , denoted n _IW(1,1) as the number of weights between the hidden layer and the output layer, n _b(1,1) as the number of hidden layer neuron thresholds, n _IW(2,1) as The number of weights between the hidden layer and the output layer, the number of n _b(2,1) output layer neuron thresholds; Step 1.2: Initialize the weight coefficient and threshold. On the basis of step 1.1, the position of the firework individual x _ik in the firework algorithm is used to represent the neuron in the neural network, and the k-th iteration process in the neural network is used. The weight coefficient between the neuron and the jth neuron

and the threshold θ _i are initialized and encoded into a vector X=[x ₁ ,x ₂ ,Λ,x _D ], and the random initialization strategy is used to initialize the vector X in the interval [-1,1], then there are weight coefficients w _ij ~ U[-1,1], Among them, i refers to the weight of the ith neuron node in the network, j refers to the weight of the jth neuron node, l represents the number of network layers where the current weight is located, and k represents the current the number of iterations; Step 1.3, calculate the error of the individual fireworks, introduce the fitness function and use the formula (1) and formula (2) to calculate the square error SSE, formula (1) and formula (2) are as follows:

Among them, t is the expected output of the network, p is the number of layers of the network, s is the number of output units of the network, and y is the output value of the network, which is as follows:

Among them, x _j is the input of the network, w _ij is the weight of the network node, θ _i is the threshold of the ith neuron in the network and θ _i = _-wi (n+1); Step 1.4, on the basis of the error of each firework individual x _i calculated in step 1.3, introduce the f _i (x) function as a fitness function, and calculate the fitness of each firework individual x _i of the vector X in step 1.2 through the fitness function. degree value, the fitness function is shown in formula (3) as follows,

Step 1.5, firework population optimization, on the basis of step 1.4, perform explosion, displacement and mutation operations for each firework individual _xi , wherein the explosion mutation operation and the Gaussian mutation mapping rule are formula (4) ~ formula (6) , h=A _i ×rand(1,-1) (4) ex _ik = x _ik +h (5) mx _ik = x _ik ×e (6) Among them, A _i is the explosion radius of the i-th firework, h is the position offset, x _ik is the k-th dimension of the i-th firework in the population, ex _ik is the spark of the i-th firework after the explosion, and mx _ik is the The Gaussian mutation spark of x _ik after Gaussian mutation, the Gaussian distribution of e～N(1,1); Step 1.6, select the next generation firework population, for the firework individual _xi after the explosion, displacement and mutation operations in step 1.5, use the formula in step 1.4 to calculate the fitness value of each firework individual _xi , and use formula (7 ) and the selection strategy of formula (8), select the optimal firework individual to form the next generation of fireworks population, and the specific selection strategy is: Select the min(f( _xi )) individual x _k with the smallest fitness value as the following firework population, and the remaining N-1 firework individuals adopt the roulette method. For the candidate individual x _i , the probability of being selected is as follows: :

Among them, R( _xi ) represents the sum of the distances between the firework individual _xi and other individuals, and the specific formula is as follows;

Step 1.7, determine the termination condition, calculate the fitness value f(x _i ) of the individual fireworks in the firework population and the Euclidean distance R(x _i ) between the individual fireworks according to formula (3) and formula (8), and determine whether the termination condition is The maximum number of iterations reached in the firework population, if it is satisfied, the minimum fitness value min(f(x _i )) of the firework individuals in the current firework population and the maximum distance between the firework individuals in the firework population max(R(x _i ) ), and take the current fireworks population as the optimal fireworks population X _best , otherwise continue to step 1.3; Step 1.8, optimize the network weights and thresholds, and use the optimal firework population X _best obtained in step 1.7 to initialize the weights and thresholds in the neural network corresponding to the vector X in step 1.2. 3. the spinning quality prediction method based on fireworks algorithm improvement BP neural network according to claim 1, is characterized in that, utilizes the data set through standardized processing in described step 3 to the FWA-BP-based method established in step 2. The specific steps for the learning and prediction of the spinning quality prediction model are as follows: Step 3.1, the training data set selection strategy, use the standardized data set in step 2.2, select 80% of the data set as the training data set, and the remaining 20% of the data set as the test data set; Step 3.2, the setting of key parameters in the firework algorithm, the size of the firework population N=70, the adjustment constant of the firework explosion radius d=5, the adjustment constant of the number of fireworks explosion sparks m=40, the upper bound value of the number of fireworks explosion sparks lm=0.8, The lower bound value of the number of fireworks explosion sparks bm=0.04, the number of Gaussian variation sparks g=5, the maximum number of iterations T=100, and the dimension of the variable D=85, which is based on step 1.1. The weight of neurons in the network model is obtained and the total number of thresholds, which are calculated by the following formula on the basis of step 2.3 D=m×s+s×n+s+n=10×7+7×1+7+1=85 Among them, m, s, n are the number of input layer neurons, hidden layer neurons and output layer neurons of the network respectively; Step 3.3, on the basis of the firework algorithm parameter setting in step 3.2, use the training data set selected in step 3.1 to train the spinning quality prediction model based on FWA-BP, and the relevant parameters in the network training process are set as, The learning rate is 0.01, the momentum factor is 0.9, the maximum number of iterations is 20000, and the minimum training error is 0.05; In step 3.4, the spinning quality prediction model based on FWA-BP is obtained through the training in steps 3.1 to 3.3, and the test data set selected in step 3.1 is used to perform test statistical analysis and experimental simulation on the prediction effect of the model.

Images