A two-stage feature selection method for polyester fiber polymerization process
By employing a two-stage feature selection method, combined with cross-correlation time delay calculation and grey relational analysis, the problems of data time delay and redundancy in the polyester fiber polymerization process were solved, enabling precise selection of key features and stable control of intrinsic viscosity, thereby improving production quality.
Patent Information
- Application Number
- CN202210421882.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-21
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2042-04-21
AI Technical Summary
Existing technologies lack theoretical guidance in the polymerization process of polyester fibers, resulting in time delays, nonlinearity, high dimensionality, and high redundancy in process data, making it difficult to accurately control intrinsic viscosity.
A two-stage feature selection method is adopted, which combines cross-correlation time delay calculation and grey relational analysis. Through preprocessing, time delay calculation, grey relational analysis and approximate Markov blanket method, key feature subsets are selected and redundant features are eliminated to achieve feature selection.
By effectively selecting key variables in the polyester fiber polymerization process, the accuracy of intrinsic viscosity control can be improved, redundancy can be reduced, and production quality stability can be enhanced.
Smart Images

Figure CN114862151B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a two-stage feature selection method for a polyester fiber polymerization process, in particular to a grey correlation analysis method based on cross-correlation time-lag calculation and a two-stage feature selection method of an approximate Markov blanket, and belongs to the field of intelligent analysis of polyester fiber data. Background Art
[0002] The textile industry is one of the pillar industries in my country, with early development and large scale. Polyester fiber is one of the important raw materials in the textile industry. Its main production process includes four parts: polyester fiber melt production, transportation, spinning and weaving processing. Among them, the melt polymerization process is the primary link in the production of polyester fiber. The intrinsic viscosity of the polyester fiber polymerization process is an important parameter to measure the quality of polyester fiber, and its stability is an important indicator to measure the production quality. In order to adjust the intrinsic viscosity of polyester fiber, the traditional method is to appropriately adjust certain parameters based on the experience of field engineers. This method often lacks relevant theoretical guidance and cannot meet the requirements of intelligent production. Summary of the invention
[0003] The purpose of the present invention is to solve the problems of time lag, nonlinearity, high dimension and high redundancy of process data caused by the industrial production mode of polyester fiber polymerization process.
[0004] In order to achieve the above object, the technical solution of the present invention is to provide a two-stage feature selection method for the polyester fiber polymerization process, characterized in that the two-stage feature selection method is a two-stage feature selection method based on a grey correlation analysis method and an approximate Markov blanket based on cross-correlation time lag calculation, comprising the following steps:
[0005] S1 obtains the process data of the polyester fiber polymerization production workshop and preprocesses the production process data. The preprocessing methods include: processing the abnormal values of the production process data; processing the noise of the production process data; processing the inconsistent dimensions of the production process data;
[0006] S2 performs time lag calculation based on cross-correlation on the production process data to determine the relationship between the process variables and the target variables in the production process data:
[0007] Assume X1(t) t=0...N is the first feature sequence data, which is the target variable, X i (t+l) t=0...N is the i-th characteristic sequence data, which is the process variable, where t and t+l represent the sampling time, N represents the maximum lag step, and the target variable X1(t) t=0...N With process variable X i (t+l) t=0...N The time lag τ i, that is, the inter-feature lag between the first feature and the i-th feature is calculated by the following formula:
[0008]
[0009] Where: T is the sampling period of the process variable;
[0010] in: is the target variable X1(t) t=0...N The mean of is the process variable X i (t+l) t=0...N The mean of
[0011] S3 introduces the time-lag relationship between features into the grey correlation analysis method considering time lag, calculates the correlation between the features and the target features, and sorts the features at the same time.
[0012] Let the candidate variable be X i =[x i (1),x i (2),…,x i (n)] is the characteristic sequence, let Y i =[y i (1),y i (2),…,y i (n)] is the target feature sequence, and the target feature sequence is further defined as a label sequence, wherein i represents the feature number and n is the sample size. Then step S3 specifically includes the following steps:
[0013] S301 performs non-quantization processing on each sequence:
[0014] X′ i =X i D i =[x i (1)d i ,x i (2)d i ,…,x i (n)d i ]=[x′ i (1),x′ i (2),…,x′ i (n)]
[0015] Y i ′=Y i D i =[y i (1)d i ,y i (2)d i ,…,yi (n)d i ]=[y′ i (1),y′ i (2),…,y′ i (n)]
[0016] In the formula, x′ i (k) = x i (k)d i =x i (k)-x i (1), y′ i (k) = y i (k)d i =y i (k)-y i (1), k = 2, ..., n; d i represents X calculated in step S2 i With Y i The time lag; D i is the starting point zeroing operator,
[0017] S302 Feature sequence X′ i The standard increment ΔS at time k i (k) is:
[0018]
[0019] S303 calculates and obtains the characteristic sequence X′ i and label sequence Y′ i The grey correlation degree γ(k,d i ), as shown below:
[0020]
[0021] In the formula,
[0022] other
[0023] S304 calculates and obtains the characteristic sequence X′ i and label sequence Y i ′’s correlation R(X′ i ,Y i ′,d i ), as shown below:
[0024]
[0025] S305 Based on the calculated R(X′ i ,Y i ′,di ) Sort all candidate variables by features, select a corresponding number of candidate variables to enter the next round of feature selection, until a feature subset consisting of the selected candidate variables is obtained;
[0026] S4 removes redundancy of the feature subset extracted in the above part using the approximate Markov blanket method based on the correlation calculated in step S3, and finally outputs the best feature subset for the polyester fiber polymerization process, thereby completing the feature selection for the polyester fiber polymerization process.
[0027] Preferably, in step S1, when processing abnormal values of production process data, the box plot method is used to calculate the lower quartile Q1, median Q2 and upper quartile Q3 of the feature data, and the quartile distance IQR=Q3-Q1 is calculated; the upper limit value of the current feature is set to Q3+1.5IQR and the lower limit value is set to Q1-1.5IQR, and the feature data exceeding the upper limit value and below the lower limit value are judged as abnormal values.
[0028] Preferably, in step S1, a sliding average method is used to process the production process data noise.
[0029] Preferably, in step S305, according to the calculated R(X′ i ,Y i ′,d i ) After all candidate variables are sorted, the top 50% of candidate variables are selected for the next round of feature selection.
[0030] Preferably, in step S4, for a feature x in the feature subset i and feature x j , feature x i is feature x j The conditions for the approximate Markov blanket are:
[0031]
[0032] In the formula, R(x i ,c,d i ) is the feature x calculated in step S3 i The correlation between the target variable and the target variable with time lag, R(x j ,c,d j ) is the feature x calculated in step S3 j The correlation between the target variable and the target variable with time lag, R(x i ,x j ,d i ) is the feature x calculated in step S3 i With feature x j The correlation between them considering time lag.
[0033] The present invention provides a two-stage feature selection method, which can effectively select features that have a greater impact on key variables from a large number of polyester fiber polymerization process variables, thereby being able to select a feature set with high correlation and low redundancy with the intrinsic viscosity in the industrial production of the polyester fiber polymerization process, so that workshop staff can more accurately control the intrinsic viscosity of the polyester fiber polymerization process. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 A schematic diagram of the two-stage feature selection process for polyester fiber polymerization process;
[0035] Figure 2 This is a schematic diagram of the polyester fiber polymerization workshop process;
[0036] Figure 3 The overall structure diagram of the two-stage feature selection method for polyester fiber polymerization process;
[0037] Figure 4 Schematic diagram of the operation logic of the two-stage feature selection method for the polyester fiber polymerization process. DETAILED DESCRIPTION
[0038] The present invention is further described below in conjunction with specific examples. It should be understood that these examples are only to help those of ordinary skill in the art understand the principles and knowledge of the present invention, and are not intended to limit the scope of the present invention, and cannot be considered to limit the application scenarios of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art may make various changes or modifications to the present invention, but the deformation, changes and conversions made to the embodiments based on the principles and purposes of the present invention also fall within the scope defined by the claims attached to this application. And it is obvious that this specification is only exemplified by preferred embodiments, and there is no need to exhaust all embodiments.
[0039] This embodiment discloses a two-stage feature selection method for a polyester fiber polymerization process, which specifically includes the following steps:
[0040] S1 obtains the process data of the polyester fiber polymerization production workshop in the production data system. The collection frequency is once every minute. The data volume for one month is collected, and a total of 43,200 data records are obtained.
[0041] The obtained production process data is preprocessed, and the preprocessing methods mainly include: processing abnormal values of production process data; processing noise of production process data; processing inconsistent dimensions of production process data. In this embodiment, the production process data is the data of each production process parameter, and each production process parameter is defined as a feature, then:
[0042] Processing of abnormal values of production process data: Use the box plot method to calculate the lower quartile Q1, median Q2 and upper quartile Q3 of the feature data, and calculate the quartile distance IQR = Q3-Q1. Set the upper limit value of the current feature to Q3+1.5IQR and the lower limit value to Q1-1.5IQR, then the feature data exceeding the upper limit value and below the lower limit value will be judged as abnormal values, such as Figure 2 shown.
[0043] Processing of production process data noise: Use the sliding average method to process data noise, as shown in the following formula:
[0044]
[0045] Where M is the smoothing coefficient, x t-n is the noisy data at the tnth moment, and y[t] represents the denoised data at the tth moment.
[0046] Processing of inconsistent data dimensions in the production process: Use maximum and minimum value normalization to process inconsistent data dimensions, as shown in the following formula:
[0047]
[0048] Where: x i and are the i-th feature data before and after normalization, x max With x min are the maximum and minimum values of the ith characteristic data before normalization, respectively. The characteristic data of the original polyester fiber polymerization are quantized to the range of 0 to 1 by a linearization method.
[0049] S2 performs time lag calculation based on cross-correlation on the production process data.
[0050] In this embodiment, the polyester fiber production method is a typical process industry production method. Changes in the previous production variables will be reflected in the final variables after a period of time. Therefore, this embodiment performs a time lag calculation based on cross-correlation on the production process data to determine the relationship between the process variables and the target variables in the production process data.
[0051] Assume X1(t) t=0...N is the first feature, which is the target variable, X i (t+l) t=0...N is the i-th feature, which is a process variable, where t and t+l represent sampling times, and N represents the maximum lag step. The target variable X1(t) t=0...N With process variable X i (t+l) t=0...N The time lag τ i, that is, the inter-feature lag between the first feature and the i-th feature can be calculated by the following formula:
[0052]
[0053] Where: T is the sampling period of the process variable;
[0054] in: is the target variable X1(t) t=0...N The mean of is the process variable X i (t+l) t=0...N The mean of .
[0055] S3 introduces the time-lag relationship between features into the grey correlation analysis method considering time lag, calculates the correlation between the features and the target features, and sorts the features at the same time.
[0056] Let the candidate variable be X i =[x i (1),x i (2),…,x i (n)] is the characteristic sequence, let Y i =[y i (1),y i (2),…,y i (n)] is the target feature sequence, and the target feature sequence is further defined as a label sequence, wherein i represents the feature number and n is the sample size. Then step S3 specifically includes the following steps:
[0057] S301 performs non-quantization processing on each sequence:
[0058] X′ i =X i D i =[x i (1)d i ,x i (2)d i ,…,x i (n)d i ]=[x′ i (1),x′ i (2),…,x′ i (n)]
[0059] Y i ′=Y i D i =[y i (1)d i ,y i (2)d i ,…,yi (n)d i ]=[y′ i (1),y′ i (2),…,y′ i (n)]
[0060] In the formula, x′ i (k) = x i (k)d i =x i (k)-x i (1), y′ i (k) = y i (k)d i =y i (k)-y i (1), k = 2, ..., n; d i represents X calculated in step S2 i With Y i The time lag; D i is the starting point zeroing operator,
[0061] S302 Feature sequence X′ i The standard increment ΔS at time k i (k) is:
[0062]
[0063] S303 calculates and obtains the characteristic sequence X′ i and label sequence Y i The grey correlation degree γ(k,d i ), as shown below:
[0064]
[0065] In the formula,
[0066] other
[0067] S304 calculates and obtains the characteristic sequence X′ i and label sequence Y i ′’s correlation R(X′ i ,Y i ′,d i ), as shown below:
[0068]
[0069] S305 Based on the calculated R(X′ i ,Y i ′,di ) All candidate variables are sorted and the top 50% of candidate variables are selected for the next round of feature selection until a feature subset consisting of the selected candidate variables is obtained.
[0070] S4 uses the approximate Markov blanket method to remove redundancy from the feature subset extracted in the above part, and finally outputs the best feature subset for the polyester fiber polymerization process, completing the feature selection for the polyester fiber polymerization process.
[0071] The data of polyester fiber polymerization process has strong redundancy. The approximate Markov blanket method is used to remove redundant features. For the feature x in the feature subset i and feature x j , feature x i is feature x j The conditions for the approximate Markov blanket are:
[0072]
[0073] In the formula, R(x i ,c,d i ) is the feature x calculated in step S3 i The correlation between the target variable and the target variable with time lag, R(x j ,c,d j ) is the feature x calculated in step S3 j The correlation between the target variable and the target variable with time lag, R(x i ,x j ,d i ) is the feature x calculated in step S3 i With feature x j The correlation between them considering time lag.
[0074] The structural framework of the above method is as follows Figure 3 As shown, the operating logic is as follows Figure 4 shown.
Claims
1. A two-stage feature selection method for polyester fiber polymerization process. It is characterized in that The two-stage feature selection method is a two-stage feature selection method based on a grey correlation analysis method and an approximate Markov blanket based on cross-correlation time lag calculation, and comprises the following steps: S1 obtains the process data of the polyester fiber polymerization production workshop and preprocesses the production process data. The preprocessing methods include: processing the abnormal values of the production process data; processing the noise of the production process data; processing the inconsistent dimensions of the production process data; S2 performs time lag calculation based on cross-correlation on the production process data to determine the relationship between the process variables and the target variables in the production process data: Assume X 1 (t) t=0...N is the first feature sequence data, which is the target variable, X i (t+l) t=0...N is the i-th characteristic sequence data, which is the process variable, where t and t+l represent the sampling time, N represents the maximum lag step, and the target variable X 1 (t) t=0...N With process variable X i (t+l) i=0...N The time lag τ i , that is, the inter-feature lag between the first feature and the i-th feature is calculated by the following formula: Where: T is the sampling period of the process variable; in: is the target variable X 1 (t) t=0...N The mean of is the process variable X i (t+l) i=0...N The mean of S3 brings the time-lag relationship between features into the grey correlation analysis method considering time lag, calculates the correlation between the features and the target features, and sorts the features at the same time; Let the candidate variable be X i =[x i (1),x i (2),…,x i (n)] is the characteristic sequence, let Y i =[y i (1),y i (2),…,y i (n)] is the target feature sequence, and the target feature sequence is further defined as a label sequence, wherein i represents the feature number and n is the sample size. Then step S3 specifically includes the following steps: S301 performs non-quantization processing on each sequence: X′ i =X i D i =[x i (1)d i ,x i (2)d i ,…,x i (n)d i ]=[x′ i (1),x′ i (2),…,x′ i (n)] Y i ′=Y i D i =[y i (1)d i ,y i (2)d i ,…,y i (n)d i ]=[y′ i (1),y′ i (2),…,y′i(n)] In the formula, x′ i (k) = x i (k)d i =x i (k)-x i (1), y′ i (k) = y i (k)d i =y i (k)-y i (1), k = 2,…,n; d i represents X calculated in step S2 i With Y i The time lag; D i is the starting point zeroing operator, S302 Feature sequence X′ i The standard increment ΔS at time k i (k) is: S303 calculates and obtains the characteristic sequence X′ i and label sequence Y i The grey correlation degree γ(k,d i ), as shown below: In the formula, other S304 calculates and obtains the characteristic sequence X′ i and label sequence Y i ′’s correlation R(X′ i ,Y i ′,d i ), as shown below: S305 Based on the calculated R(X′ i ,Y′ i ,d i ) Sort all candidate variables by features, select a corresponding number of candidate variables to enter the next round of feature selection, until a feature subset consisting of the selected candidate variables is obtained; S4 removes the redundancy of the feature subset extracted in the above part by using the approximate Markov blanket method based on the correlation calculated in step S3, and finally outputs the best feature subset for the polyester fiber polymerization process, completing the feature selection for the polyester fiber polymerization process. Among them, for the feature x in the feature subset i and feature x j , feature x i is feature x j The conditions for the approximate Markov blanket are: In the formula, R(x i ,c,d i ) is the feature x calculated in step S3 i The correlation between the target variable and the target variable with time lag, R(x j ,c,d j ) is the feature x calculated in step S3 j The correlation between the target variable and the target variable with time lag, R(x i ,x j ,d i ) is the feature x calculated in step S3 i With feature x j The correlation between them considering time lag.
2. A two-stage feature selection method for polyester fiber polymerization process as claimed in claim 1, It is characterized in that In step S1, when processing the abnormal values of the production process data, the box plot method is used to calculate the lower quartile Q1, median Q2 and upper quartile Q3 of the feature data, and the quartile distance IQR = Q3-Q1; the upper limit value of the current feature is set to Q3+1.5IQR and the lower limit value is Q1-1.5IQR, and the feature data exceeding the upper limit value and below the lower limit value are judged as abnormal values.
3. A two-stage feature selection method for polyester fiber polymerization process as claimed in claim 1, It is characterized in that In step S1, the sliding average method is used to process the noise of the production process data.
4. A two-stage feature selection method for polyester fiber polymerization process as claimed in claim 1, It is characterized in that In step S305, according to the calculated R(X′ i ,Y i ′,d i ) After all candidate variables are sorted, the top 50% of candidate variables are selected for the next round of feature selection.
