A method for predicting resistance of a cigarette filter rod forming machine production process and a storage medium

Abstract

The application discloses a filter rod forming machine production process resistance prediction method and a storage medium, including using improved entropy PCA to reduce the dimension of the forming machine operation parameter data set; using the improved firefly optimization algorithm FA to adjust the Elman neural network threshold to make up for the deficiency of being easy to fall into local optimum; applying the trained optimal weight to the prediction model to predict the resistance of the filter rod forming machine at the next time. The application is based on the massive production operation and detection information contained in the filter rod forming machine in the cigarette factory, and adopts an artificial intelligence method to train and learn the filter rod forming machine production information, and further proposes a filter rod forming resistance prediction method based on PCA-FA-Elman neural network to guide the advanced adjustment of the resistance control parameter of the operator and achieve the purpose of reducing the filter rod defective rate. The prediction method provided by the application provides an important reference method for controlling the filter rod resistance quality parameter.

Description

TECHNICAL FIELD

[0001] The present application relates to a cigarette filter rod forming resistance prediction method, in particular to a cigarette filter rod forming machine production process resistance prediction method based on PCA-FA-Elman neural network and a storage medium. BACKGROUND

[0002] Since the resistance of the filter rod directly affects the experience of the consumer smoking the cigarette and the quality of the full combustion of the cigarette, the prediction control of the resistance, as a key quality parameter of the filter rod, is very important. The manual offline sampling inspection of the resistance quality of the filter rod forming machine generally has problems such as low efficiency, offline detection lag of the quality test table, and cannot well meet the current production requirements.

[0003] The filter rod resistance prediction can guide the machine operator to intervene in advance on each control parameter, stabilize the production index, reduce the defective rate, thereby reduce the rejection amount, and thus effectively improve the economic benefit of the filter rod production, and can replace the manual offline sampling inspection to reduce the labor intensity. SUMMARY

[0004] The present application aims to solve the problems of low efficiency of manual offline sampling inspection of the resistance quality of the filter rod forming machine, offline detection lag of the quality test table, and proposes a filter rod forming machine production process resistance prediction method to improve the intelligent real-time control level of the filter rod production, thereby improving the qualified rate of the filter rod.

[0005] In order to achieve the above effect, the technical scheme of the present application is:

[0006] A cigarette filter rod forming machine production process resistance prediction method, by using the PCA algorithm based on the improved local entropy, reducing the input dimension of the Elman neural network, and using the FA optimized Elman neural network after introducing the filling function without parameters, improving the prediction model accuracy, thereby more accurately predicting the resistance data, to improve the quality of the filter rod and reduce the defective rate. Specifically includes the following steps:

[0007] Step 1, the improved local entropy algorithm is used to reduce the dimension of the forming machine operation parameter data set;

[0008] Step 1.1, the local entropy of the filter rod forming machine historical production data is improved;

[0009] Step 1.2, the data after the improved local entropy is used as the input of the PCA algorithm, and the standardized data set, the covariance matrix of the standardized data set, and the eigenvalue and corresponding eigenvector of the covariance matrix are calculated;

[0010] For the input of the system (stabilizing roller pressure, opening ratio, speed ratio, thread roller pressure, plasticizer spraying rate, etc.), the sample set is represented as X=[X1,X2,...,Xn ], n is the sample serial number.

[0011] The Shannon entropy used in the application is defined as follows:

[0012]

[0013] Where p(x) is the probability density at x. Usually, the probability density estimation is used instead of the probability density which cannot be actually obtained to calculate.

[0014] The kernel density estimation is used in the probability density estimation:

[0015]

[0016] The generalized uniform Chebyshev polynomial kernel function (GUCK) is used in the kernel density estimation in the application, which is a vector form of kernel density, and its expression is as follows:

[0017]

[0018] The filter rod forming machine can have multiple different operation modes in the production process, and different operation modes need to use different data processing methods. The corresponding kernel density can better perform subsequent prediction. In order to improve the accuracy, the kernel density estimation function with the KNN idea is cited in the application. As follows:

[0019]

[0020] h = [h1, h2, h3,..., hn] d ]

[0021] Where i is the sampling time, h is the window width, d is the space dimension, and n is the number of observers.

[0022]

[0023] Step 1.3, the number of principal components is calculated according to the following formula:

[0024]

[0025] Step 1.4: arrange the eigenvalues of the covariance calculated from large to small, select the first l elements as the principal components, and sequentially form a matrix with the unit eigenvectors corresponding to the principal components. Multiply the original data set to obtain the final reduced data set.

[0026] Step 2: optimize the Elman network using the improved FA, and train the neural network using the reduced data set.

[0027] Step2.1: The reduced dimension data set obtained in Step1 is used as the network model training input set;

[0028] Step2.2: The error function between the predicted value and the actual value of the Elman neural network is calculated, and the minimum value is taken as the target.

[0029] Step2.3: Update the firefly position, update the light intensity ranking, and select the optimal solution;

[0030] Step2.4: Determine whether the maximum number of iterations is reached, if not, go to Step2.2, if reached, select the optimal solution as the weight value of each layer of the Elman neural network;

[0031] Step3: Use the optimized prediction model to predict the suction resistance of the filter rod forming machine.

[0032] A computer readable storage medium having a computer program stored thereon, the computer program being executable by a processor to implement the steps of the suction resistance prediction method of the cigarette filter rod forming machine production process according to the present application.

[0033] The beneficial effects of the present application are:

[0034] The present application reduces the input dimension of the Elman neural network based on the improved local entropy PCA algorithm, and optimizes the Elman neural network using the FA with the introduction of the parameter-free filling function, improves the prediction model accuracy, and thus more accurately predicts the suction resistance data, to improve the filter rod quality and reduce the rate of defective products. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The overall step flowchart of the method of the present application.

[0036] Figure 2 The algorithm flowchart of the method of the present application. DETAILED DESCRIPTION

[0037] As shown in Figure 1 and Figure 2 A suction resistance prediction method for a cigarette filter rod forming machine production process based on PCA-FA-Elman neural network, which reduces the data dimension as the input of the prediction model using the improved local entropy PCA algorithm. A parameter-free filling function is introduced in the FA, and the error between the predicted value and the actual value in the model training is used as the objective function to reduce the probability of the FA falling into local optimum. The improved FA is used to optimize the Elman neural network to obtain the optimal weight value, and the suction resistance of the filter rod forming is predicted.

[0038] The filter rod suction resistance produced by the forming machine is directly related to the fiber material specification and the machine control parameter setting, and the application provides a filter rod forming suction resistance prediction method based on a PA-FA-Elman neural network, which can overcome the hysteresis of off-line detection of suction resistance, guide the operator to change the control quantity in real time, and improve the filter rod qualification rate.

[0039] A historical data set of a filter rod forming machine is acquired, and is converted into an original observation data matrix X.If the local probability density estimation matrix of the observation data X is P=R m×n , n is the sample quantity, and m is the variable.L is the moving window width, and the moving window P t of the t moment is as follows:

[0040]

[0041] At this time, the local entropy variable h t is as follows:

[0042] h t =[h1(t)h2(t)...h m (t)] T

[0043]

[0044] According to the above steps, the original data matrix subjected to the local entropy transformation processing becomes a new data matrix A m×n .

[0045] The matrix A is subjected to dimension reduction processing as a new sample set.Each row of the matrix A is subtracted by the average value of the row; the matrix B=1 / m(AA T ) is calculated; the eigenvalues of the covariance matrix B and the eigenvectors corresponding to each eigenvalue are calculated; the eigenvectors are arranged from top to bottom in a row to form a matrix according to the size of the corresponding eigenvalues, and the first k rows constitute a new matrix M; Z=MA is the data after the dimension reduction to k dimensions.

[0046] According to the above steps, the m variables in the historical data set of the filter rod forming machine are reduced to k main variables, the data dimension is reduced, and the data after the dimension reduction has the independent characteristics and is used as the input data.

[0047] The topological structure of the Elman neural network is determined according to the matrix Z.

[0048] The weight values of each layer are initialized; the sample values are inputted and the output values of each layer are calculated; the target function is obtained according to the error principle; the glowworm population is initialized;

[0049] If the actual output of the Elman neural network at the k step is y dLet E(k) be the error function, as shown below:

[0050]

[0051] Introducing a parameterless filling function allows the objective function to escape local optima.

[0052] Choose an initial point x1∈Ω, and starting from x1, use the local minimization algorithm to make f(x) (in this invention, the error function value) obtain an initial local minimum point x1. * ;

[0053] Select an initial sequence {x k+1 i |i = 1, 2, ..., m}, such that for some δ k >0, there is x k+1 i ∈Ω, k:=1; Let i=1;

[0054] If i > m, then it is impossible to find a value greater than f(x) within the current range. * For smaller points, take x. * The algorithm terminates when the point is the global minimum.

[0055] If i ≤ m, let x := x k+1 i If f(x) > f(x) * If ), then a parameterless fill function is introduced:

[0056]

[0057]

[0058] Minimize P{x,x k * Given a point x, find a new point x; if f(x) ∈ {x, x}. <f(x * Starting from x, use the local minima algorithm to find the local minimum point x of the objective function f(x). k * And satisfy f(x) k+1 * ) <f(x k * ).

[0059] Update firefly positions; update light intensity; reorder fireflies.

[0060] The position update function is as follows, indicating that the low-brightness firefly moves towards the brighter firefly:

[0061]

[0062] wherein, represents the current individual position, is the attraction degree of the firefly, and aε is a random term, and β0represents the maximum attraction degree when the distance between two fireflies is r=0. The firefly brightness update function is as follows:

[0063]

[0064] wherein, I0is the light intensity when the distance between the firefly and other individuals is r=0, i.e. the self brightness, γ is the attraction coefficient, and r is the current distance between two fireflies.

[0065] selecting the current optimal solution;

[0066] if the maximum iteration number has not been reached, the improved FA algorithm is continuously used to optimize the Elman neural network to obtain the optimal solution;

[0067] if the maximum iteration number is reached, the optimal solution is taken as the latest weight value of the Elman neural network.

[0068] In summary, the Elman neural network optimized by the improved FA is trained by using all data of the training set, when the error function value reaches an ideal state or a set state, the trained Elman neural network is used to calculate the suction resistance parameter of the filter rod forming machine production process at the next time, to obtain a prediction result, to guide the operator to change the control parameter in real time, and to realize the suction resistance overrun alarm.

[0069] The above description is one specific embodiment of the present application, but the present application is not limited to the above embodiment, and various changes can be made within the scope of the knowledge of those skilled in the art without departing from the purpose of the present application.

Claims

1. A method of predicting resistance to draw of a cigarette filter rod produced by a cigarette filter rod making machine, characterized by, The steps comprise: Step 1: using the improved entropy principal component analysis method PCA to reduce the dimension of the filter rod forming machine suction resistance related historical data and taking the same as the Elman neural network sample data; Step 2: determining the network topology structure and performing network initialization, optimizing the Elman neural network by using the improved FA, and training the optimized Elman neural network; Step 3: using the trained Elman neural network to predict the suction resistance of the filter rod forming machine; The Step 1 comprises: Step 1.1: improving the local entropy of the filter rod forming machine historical production data; Step 1.2: taking the data after improving the local entropy as the input of the PCA algorithm, standardizing the same, calculating the covariance matrix of the standardized data set, and the eigenvalue and corresponding eigenvector of the covariance matrix; The input sample set of the filter rod maker suction resistance related historical data is represented as wherein n is the sample serial number; Entropy is defined as: where p(x) is the probability density at x. Using kernel density estimation in probability density estimation: ​ Using the generalized uniform Chebyshev polynomial kernel function GUCK in kernel density estimation: ； The kernel density estimation function with the KNN idea: ； ； where: i is the sampling time, h is the window width, d is the spatial dimension, and n is the number of observers; ​ Step 1.3: the number of principal components is calculated according to the following formula: ； Step 1.4: The eigenvalues of the covariance are sorted in descending order, and the first elements are selected as the principal components. The unit eigenvectors corresponding to the principal components are sequentially arranged to form a matrix, which is multiplied by the original data set to obtain the final reduced dimension data set.

2. The method of claim 1, wherein, The Step 2 comprises: Step 2.1: taking the dimension-reduced data set obtained from Step 1 as the neural network model training input set; Step 2.2: taking the error function between the predicted value and the true value of the Elman neural network as the target, when the target function value f(x) falls into a local optimum, introducing a filling function without parameters to make it jump out of the neighborhood of the current local minimum point, and finding another better local minimum point from the current local minimum point; Step 2.3: updating the firefly position, updating the light intensity ranking, and selecting the optimal solution; Step 2.4: determining whether the maximum number of iterations is reached, if not, going to Step 2.2, if yes, selecting the optimal solution as the weight value of each layer of the Elman neural network.

3. The method according to claim 1, characterized in that: The filter rod forming machine suction resistance related historical data comprises a training data set composed of stable roller pressure, opening ratio, return speed ratio, thread roller pressure, plasticizer spraying rate, and artificial offline sampling and quality inspection station offline sampling suction resistance.

4. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program can be executed by a processor to realize the steps of the suction resistance prediction method of the cigarette filter rod forming machine production process according to any one of claims 1-3.

Images