Process parameter time sequence registration method based on time sequence correlation matrix and improved particle swarm algorithm

Abstract

The application discloses a process parameter time sequence registration method based on a time sequence correlation matrix and an improved particle swarm algorithm, and comprises the following steps: preprocessing time sequence data of each batch process to obtain pretreated time sequence data of each batch process; constructing a regularity feature screening method based on a time sequence correlation matrix to screen a process parameter set which is significantly affected by time delay; obtaining a screened process parameter time sequence correlation matrix according to the screened process parameter set; optimizing the screened process parameter time sequence correlation matrix according to the improved particle swarm algorithm to find an optimal lag time solution set of the process parameter; and according to the optimal lag time solution set of the process parameter, accurately aligning each process parameter time sequence which is significantly affected by time delay with a quality index in a time sequence by shifting the time sequence backward by the corresponding optimal lag time to obtain registered data. The application provides important support for process manufacturing enterprises to guarantee process quality and improve processing efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to a process parameter time sequence registration method based on a time sequence correlation matrix and an improved particle swarm algorithm, and belongs to the technical field of data processing. BACKGROUND

[0002] Process industry is an indispensable part of modern manufacturing industry and an important pillar industry of current national economy and social development. In actual process industrial production process, process data is usually collected by sensors in real time and recorded synchronously according to fixed time intervals to form continuous time series data. However, due to the complex physical and chemical reactions involved in the production process, the change of process parameters often cannot be immediately reflected in the quality index, thereby causing time delay effect between the two. This delay effect causes the input of the production system to be unable to be timely reflected in the output, resulting in that the decision is based on distorted data, which increases the potential risk. Therefore, effectively eliminating the time delay effect is crucial to improve the accuracy and reliability of production decision. SUMMARY

[0003] The present application provides a process parameter time sequence registration method based on a time sequence correlation matrix and an improved particle swarm algorithm to solve the time delay effect existing in process manufacturing.

[0004] The technical scheme of the present application is:

[0005] According to the first aspect of the present application, a process parameter time sequence registration method based on a time sequence correlation matrix and an improved particle swarm algorithm is provided, and the process parameter time sequence registration method based on a time sequence correlation matrix and an improved particle swarm algorithm is characterized by comprising the following steps:

[0006] S1, obtaining process technology time series data of different batches of process production lines; pre-processing each batch of process technology time series data to obtain pre-processed process technology time series data of each batch; wherein the pre-processed process technology time series data is constructed by M process parameters and a quality index;

[0007] S2, constructing a regularity feature screening method based on a time sequence correlation matrix to obtain a process parameter time sequence correlation matrix, and identifying and screening out a process parameter set significantly affected by time delay according to the time sequence correlation matrix of the process parameters;

[0008] S3, obtaining a screened process parameter time sequence correlation matrix according to the screened process parameter set;

[0009] S4, optimizing the screened process parameter time sequence correlation matrix according to the improved particle swarm algorithm to find an optimal lag time solution set of the process parameters;

[0010] S5. Based on the optimal lag time solution set of process parameters, shift the time series of each process parameter that is significantly affected by time delay backward to the corresponding optimal lag time to achieve accurate alignment with the quality index in the time series, and obtain the registered data.

[0011] Further, S2 includes:

[0012] S2.1 Constructing the time-series correlation matrix of process parameters: Calculate the cross-correlation coefficients of the time series relative quality indicators of each process parameter in the pre-processed process time series data of each batch at different lag times, thereby constructing the time-series correlation matrix of process parameters;

[0013] S2.2. Use the regularity feature screening method to analyze and quantify the time-series correlation matrix of process parameters, thereby screening out the set of process parameters that are significantly affected by the time delay effect.

[0014] Furthermore, the regularity feature screening method includes: obtaining the time delay correlation coefficient of each process parameter under different lag times based on the time series correlation matrix of the process parameters; obtaining the average time delay correlation coefficient of each process parameter based on the time delay correlation coefficient of each process parameter under different lag times; judging based on the average time delay correlation coefficient of each process parameter and a preset threshold: retaining the current process parameter if the average time delay correlation coefficient of the current process parameter is less than the preset threshold; and obtaining a set of process parameters based on the retained process parameters.

[0015] Furthermore, the expressions for the time delay correlation coefficients of each process parameter under different lag times and the average time delay correlation coefficients of each process parameter are as follows:

[0016]

[0017] Among them, G i (t l ) represents the time lag as t l The time delay correlation coefficient of the i-th process parameter per second; N represents the total batch; f i,q (t l ) represents the time lag as t l The cross-correlation coefficient corresponding to the i-th process parameter of the q-th batch per second. The lag time is t l The average cross-correlation coefficient of the i-th process parameter per second; G i Let be the average time delay correlation coefficient of the i-th process parameter.

[0018] Further, the correlation coefficient between the screened process parameters and quality indicators data significantly affected by time delay is used as fitness, the maximum correlation coefficient is used as target, and the improved particle swarm optimization algorithm is used for global optimization to search for the global optimal lag time, and then the optimal lag time solution set of the process parameter time sequence is obtained.

[0019] According to a second aspect of the present application, a process parameter time sequence registration system based on a time sequence correlation matrix and an improved particle swarm algorithm is provided, which comprises a module of the method in any one of the above.

[0020] According to a third aspect of the present application, a processor is provided, which is used to run a program, wherein the process parameter time sequence registration method based on a time sequence correlation matrix and an improved particle swarm algorithm in any one of the above is executed when the program is run.

[0021] According to a fourth aspect of the present application, a computer readable storage medium is provided, which comprises a stored program, wherein the process parameter time sequence registration method based on a time sequence correlation matrix and an improved particle swarm algorithm in any one of the above is executed when the program is run.

[0022] The present application has the following beneficial effects: the process parameters affected by time delay effect are identified and screened by the regularity feature screening method based on a time sequence correlation matrix, so that the effective extraction of key parameters is ensured; further, the improved particle swarm algorithm is used to eliminate the influence of process error and reduce the complexity of the time sequence correlation matrix, so that the optimal lag time solution set is quickly found, and the accurate alignment of the process parameters and quality indicators in the time sequence is realized. In addition, the feasibility and effectiveness of the method in the time sequence registration are proved by the verification of the data of a certain process production line, which provides important support for process manufacturing enterprises to ensure process quality and improve processing efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0023] Fig. 1 The figure is a flow structure diagram of the present application;

[0024] Fig. 2 The figure is a regularity feature screening method structure diagram based on a time sequence correlation matrix of the present application;

[0025] Fig. 3 The figure is a structure diagram based on an improved particle swarm algorithm of the present application. DETAILED DESCRIPTION

[0026] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other in any manner without conflict.

[0027] Embodiment 1: As shown, according to the first aspect of the embodiments of the present application, a process parameter time sequence registration method based on a time sequence correlation matrix and an improved particle swarm algorithm is provided, comprising the following steps: Figs. 1-3

[0028] S1, obtaining process time sequence data of different batches of processes of a production line; preprocessing the process time sequence data of each batch to obtain preprocessed process time sequence data of each batch; wherein the preprocessed process time sequence data is constructed by M process parameters and a quality index; it should be noted that the preprocessing is to remove the head and tail operations, that is, to retain the data between the critical positions of the head and tail;

[0029] S2, constructing a regularity feature screening method based on a time sequence correlation matrix, to obtain a process parameter time sequence correlation matrix, and identifying and screening a set of process parameters significantly affected by time delay according to the time sequence correlation matrix of the process parameters;

[0030] S3, obtaining a screened process parameter time sequence correlation matrix according to the screened process parameter set;

[0031] S4, optimizing the screened process parameter time sequence correlation matrix according to the improved particle swarm algorithm to find an optimal lag time solution set of the process parameters; that is, using the improved particle swarm algorithm to obtain the optimal lag time of each process parameter, and establishing an optimal lag time solution set of the process parameters according to the optimal lag time of each process parameter, that is, a set of optimal lag times of all process parameters.

[0032] S5, according to the optimal lag time solution set of the process parameters, shifting the time sequence of each process parameter significantly affected by time delay backward by the corresponding optimal lag time to realize accurate alignment with the quality index in the time sequence, and obtaining registered data.

[0033] Further, the registered data is registered, and the missing sequence / redundant sequence is removed based on the quality index to obtain a registration sequence; or the registered data is filled in using an LSTM preprocessing model.

[0034] Further, the S2 comprises: ​

[0035] S2.1, construct process parameter time series correlation matrix: calculate the time series relative quality indicators of each process parameter in the pretreated process time series data of each batch, and the cross-correlation coefficient of different lag time, so as to construct the time series correlation matrix of process parameters; the time series relative quality indicators of process parameters are moved back h seconds (in the embodiment of the application, h = 1, 2,..., 360, which means lagging h seconds), and the expression of the cross-correlation coefficient R(h) is as follows:

[0036]

[0037] Wherein, represents the mth process parameter of the nth batch collected at t time point moving back h seconds, represents the quality indicator of the nth batch collected at t time point; represents the joint expectation of represents the expectation of represents the expectation of represents the standard deviation of represents the standard deviation of It should be noted that when h is 0s, it means no shift, that is, the time series of process parameters and quality indicators are in the same position;

[0038] As shown in Fig. 2 , the original time series represents the original process parameter time series of a batch, X1, X2,..., X M represents the first, second, and Mth process parameter time series, Y represents the quality indicator time series, and the cross-correlation coefficient of the first process parameter time series relative to the quality indicator at different lag times is used for illustration, that is, the first process parameter time series represented by the first column is moved down by the required time (that is, the lag time), and then the cross-correlation coefficient of the moved first process parameter time series relative to the quality indicator is calculated, and the cross-correlation coefficient of the moved second to Mth process parameter time series relative to the quality indicator is calculated, so as to construct the time series correlation matrix of process parameters; for example, the expression is as follows:

[0039]

[0040] Wherein, f 1,n (t T ) is the cross-correlation coefficient of the first process parameter of the nth batch corresponding to the lag time t T second; f M,n (t T ) is the cross-correlation coefficient of the first process parameter of the nth batch corresponding to the lag time t T ​​​​the correlation coefficient corresponding to the Mth process parameter of the nth batch at t seconds; in the above matrix, the time sequence correlation matrix of the 1st process parameter is from the 1st column to the Tth column, and the time sequence correlation matrix of the 2nd process parameter to the Mth process parameter is in turn.

[0041] S2.2, using the regularity feature screening method to analyze the time sequence correlation matrix of the quantitative process parameter, so as to screen out a process parameter set significantly affected by the time delay effect. The present application screens out the process parameter set significantly affected by the time delay effect through the regularity feature screening method, which helps to remove the irrelevant factors in the process data which are irrelevant to the processing time delay effect.

[0042] Further, the regularity feature screening method comprises:

[0043] According to the time sequence correlation matrix of the process parameter, the time delay correlation coefficient of each process parameter under different lag time is obtained;

[0044] According to the time delay correlation coefficient of each process parameter under different lag time, the average time delay correlation coefficient of each process parameter is obtained;

[0045] According to the average time delay correlation coefficient of each process parameter and the preset threshold value, the current process parameter is retained under the condition that the average time delay correlation coefficient of the current process parameter is less than the preset threshold value; and the process parameter set is obtained according to the retained process parameter. In the embodiment of the present application, the preset threshold value is set to 0.041887, that is, the process parameter corresponding to the average time delay correlation coefficient of the process parameter greater than or equal to 0.041887 is removed; the preset threshold value in the present application adopts a relatively large value, based on which the process parameter affected by the time delay effect to a large extent can be retained to the greatest extent, so as to avoid the process parameter affected by the time delay effect to a large extent being removed which will have a bad influence on the subsequent steps.

[0046] Further, the time delay correlation coefficient of each process parameter under different lag time and the average time delay correlation coefficient of each process parameter are expressed as follows:

[0047]

[0048] Wherein, G i (t l ) is the time delay correlation coefficient of the i th process parameter at t l seconds; N represents the total batch; f i,q (t l ) is the correlation coefficient corresponding to the i th process parameter of the q th batch at t l seconds, is the correlation coefficient corresponding to the i th process parameter of the q th batch at t lthe average value of the cross-correlation coefficient of the i th process parameter at t seconds; l = 1, 2,..., T, t T represents the maximum lag time; G i is the average time-delay correlation coefficient of the i th process parameter; in the embodiment of the present application, t 1 = 1 second, t T = 360 seconds, that is, 360 seconds is the maximum lag value of the process parameter time series; taking the sheet drying process as an example, the process flow time from input to output of raw materials is generally 5-6 minutes, so the present application sets 360 seconds as the maximum lag value, which conforms to the lag law in the actual production process.

[0049] Further, between step S1 and step S2, normalization processing is further included, that is, the pre-processed process flow time series data of each batch is normalized to accelerate the calculation efficiency.

[0050] The improvement point of the improved particle swarm optimization algorithm is that the inertia weight linearly decreases, the acceleration factor self-adapts, and the particle position is limited; the cross-correlation coefficient of the screened process parameter and quality index data significantly affected by time delay is used as the fitness, the maximum cross-correlation coefficient is used as the target, the improved particle swarm optimization algorithm is used for global optimization, the global optimal lag time is searched, and then the optimal lag time solution set of the process parameter time series is obtained.

[0051] As shown in Fig. 3 , the specific steps are as follows:

[0052] S4.1, randomly initialize the particle swarm: the size of the particle swarm is set to h particles, the dimension of each particle is 1, and each particle in the particle swarm represents a lag time solution; the initial position of the particle is randomly generated in the search space [0, t T ]; the speed vector is initialized to 0; in the embodiment of the present application, h is 12, and the lag time range of the process parameter is set to [0, 360], which is used to capture the correlation between the process parameter and the quality index under different lag times;

[0053] S4.2, calculate the fitness of each particle; the fitness value is determined by the cross-correlation coefficient of the screened process parameter and quality index data significantly affected by time delay;

[0054] S4.3, update the individual optimal position and global optimal position of the particle;

[0055] S4.4, update the inertia weight according to the inertia weight linearly decreasing strategy, and update the acceleration factor according to the acceleration factor self-adapting strategy;

[0056] S4.5, update the speed and position of the particle;

[0057] S4.6, Strategy of limiting particle position: used to check whether there is a particle whose position exceeds the search range, and put the particle back into the search range if it exceeds the search range;

[0058] S4.7, Judgment of whether the set termination condition is reached: if the termination condition is reached, output the optimal lag time; if not, repeat steps S4.2-S4.7 until the termination condition is reached;

[0059] S4.8, Judgment of whether the process parameter upper limit is reached, if the maximum iteration number is reached, output the optimal lag time, if the maximum iteration number is not reached, repeat steps S4.2-S4.8.

[0060] The expression of the inertia weight linearly decreasing strategy is:

[0061]

[0062] In the formula, ωtis the inertia weight at the tthiteration, ωmaxis the maximum inertia weight, ωminis the minimum inertia weight, and Tmaxis the maximum iteration number. (t) max min

[0063] The expression of the adaptive change strategy of the acceleration factor is:

[0064]

[0065] In the formula, ctis the individual and global acceleration factor when the tthiteration is performed, cmaxis the individual and global maximum acceleration factor, and cminis the individual and global minimum acceleration factor. 1max 2max 1min 2min Tmaxis the maximum iteration number.

[0066] The expression of the search strategy in the strategy of limiting particle position is:

[0067] 0.1h min ≤h i ≤0.1h max

[0068] In the formula, h represents the particle position, h is the lower limit of particle search, and h is the upper limit of particle search. i min max

[0069] Further, in the above, the inertia weight decreases from 0.9 to 0.2, gradually reducing the dependence of the particle on the previous speed and enhancing the convergence ability, i.e., ωt= 0.9-0.7t. max min ​​​​​​​​​​​Take 0.2; individual learning coefficient: from 2 to 1, so that each particle gradually reduces the dependence on its own historical optimal position in the search process, that is, c 1max Take 2, c 1min Take 1; global learning coefficient: from 1 to 2, gradually enhancing the ability of particles to follow the global optimal position, promoting the exploration of the global optimal solution by the colony, that is, c 2max Take 2, c 2min Take 1; the algorithm sets Tmax to 100.

[0070] Exemplarily, an optional embodiment of the present application is described in detail below, but the content of the present application is not limited to the scope described.

[0071] This embodiment aims at the problems of high coupling correlation degree of process data, strong sample time sequence, and time delay effect between process parameters and quality indicators, and proposes a process parameter time sequence registration method based on time sequence correlation matrix and improved particle swarm algorithm. First, the time sequence correlation matrix is constructed using the regularity feature screening method based on the time sequence correlation matrix, and the process parameter set significantly affected by time delay is screened out; then, the improved particle swarm algorithm is used to optimize the screened time sequence correlation matrix, so as to find the optimal lag time solution set of the process parameters, and realize the accurate alignment of the process parameters and the quality indicators in the time sequence.

[0072] The sample data set comes from a thin plate drying process production line of a certain process manufacturing enterprise. The process data of October 2022 derived from the process production line is obtained, and the above registration method is used to normalize the obtained process data, construct the time sequence correlation matrix, screen the regularity features, improve the particle swarm optimization, and perform data registration operation to obtain the registered data. Then, the registered data is filled according to the LSTM model, and the training set and test set are divided according to the ratio of 7:3. At the same time, a variety of machine learning regression models are used for training, and the feasibility and effectiveness of the present application are verified according to the test set. The results are shown in Table 1.

[0073] Table 1: Prediction results of each model

[0074]

[0075] In this embodiment, the model evaluation indexes used include mean absolute error (MAE), mean square error (MSE), and goodness of fit (R 2 ).

[0076]

[0077] In the formula: s is the number of test set samples, y 1,j is the predicted quality indicator value, and y 2,jthe actual quality index value, the average value of the actual quality index data.

[0078] From Table 1, it can be seen that the fitting goodness and error of different prediction models of the data aligned in time sequence by the method of the application are significantly better than those of the data not aligned in time sequence. This shows that the time delay effect can effectively improve the accuracy of the prediction model. In addition, these results further verify that the process data aligned using the method of the application can more accurately reflect the causal relationship between the model input and the output, further improving the quality of the modeling data.

[0079] According to a second aspect of the embodiments of the application, a process parameter time sequence alignment system based on a time sequence correlation matrix and an improved particle swarm algorithm is provided, which comprises a first module for performing S1, acquiring process time sequence data of different batches of a process production line; preprocessing each batch of process time sequence data to obtain preprocessed process time sequence data of each batch; wherein the preprocessed process time sequence data is constructed by M process parameters and a quality index; a second module for performing S2, constructing a regularity feature screening method based on a time sequence correlation matrix, to obtain a process parameter time sequence correlation matrix, and identifying and screening a set of process parameters significantly affected by time delay according to the time sequence correlation matrix of the process parameters; a third module for performing S3, obtaining a screened process parameter time sequence correlation matrix according to the screened set of process parameters; a fourth module for performing S4, optimizing the screened process parameter time sequence correlation matrix according to the improved particle swarm algorithm, and finding an optimal lag time solution set of the process parameters; and a fifth module for performing S5, precisely aligning each process parameter time sequence significantly affected by time delay with the quality index in time sequence by shifting the time sequence backward by the corresponding optimal lag time, to obtain aligned data. The term "module" as used above can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the above embodiments is preferably implemented in software, hardware, or a combination of software and hardware is also possible and contemplated. For parts not described in detail for each module, please refer to the related description of the embodiments.

[0080] According to a third aspect of the embodiments of the application, a processor is provided, which is used to run a program, wherein the program performs the process parameter time sequence alignment method based on a time sequence correlation matrix and an improved particle swarm algorithm according to any one of the above embodiments when running.

[0081] According to a fourth aspect of the embodiment of the present application, a computer readable storage medium is provided, which includes a stored program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute the process parameter timing registration method based on the timing correlation matrix and the improved particle swarm algorithm according to any one of the above embodiments when the program is running.

[0082] In an exemplary embodiment, the computer readable storage medium can include, but is not limited to, a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0083] The specific embodiments of the present application are described in detail above with reference to the accompanying drawings, but the present application is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present application.

Claims

1. A method for process parameter timing registration based on timing correlation matrix and improved particle swarm optimization algorithm, characterized in that, The method comprises the following steps: S1, obtaining process production line different batches of process time series data; S2, constructing a regularity feature screening method based on a time sequence correlation matrix to obtain a process parameter time sequence correlation matrix, and identifying and screening a process parameter set significantly affected by time delay according to the process parameter time sequence correlation matrix; S3, obtaining a screened process parameter time sequence correlation matrix according to the screened process parameter set; S4, optimizing the screened process parameter time sequence correlation matrix according to the improved particle swarm optimization algorithm to find an optimal lag time solution set of the process parameters; S5, aligning the time series of the screened process parameters significantly affected by time delay with the quality index in time series by backward shifting the corresponding optimal lag time to obtain registered data; The S2 comprises: S2.1, constructing a process parameter time sequence correlation matrix: calculating the cross-correlation coefficients of the time series of each process parameter in the pre-processed process time series data of each batch with the quality index at different lag times, thereby constructing a time sequence correlation matrix of the process parameters; S2.2, using the regularity feature screening method to analyze and quantify the time sequence correlation matrix of the process parameters, thereby screening a process parameter set significantly affected by time delay effect; The regularity feature screening method comprises: According to the time sequence correlation matrix of the process parameters, obtain the time delay correlation coefficients of each process parameter at different lag times; According to the time delay correlation coefficients of each process parameter at different lag times, obtain the average time delay correlation coefficients of each process parameter; According to the average time delay correlation coefficients of each process parameter and the preset threshold, if the average time delay correlation coefficients of the current process parameter are less than the preset threshold, the current process parameter is retained; and according to the retained process parameters, a process parameter set is obtained. The time delay correlation coefficients of each process parameter at different lag times and the average time delay correlation coefficients of each process parameter are expressed as follows:

2. The method according to claim 1, wherein, The cross-correlation coefficients of the screened process parameters and the quality index data significantly affected by time delay are used as fitness, and the maximum cross-correlation coefficient is used as the target, and the improved particle swarm optimization algorithm is used for global optimization to search for the global optimal lag time, and then the optimal lag time solution set of the process parameter time sequence is obtained. ； ； wherein, is the time-lag correlation coefficient of the i th process parameter when the lag time is seconds; N represents the total batch number; is the cross-correlation coefficient corresponding to the i th process parameter of the q th batch when the lag time is seconds, is the cross-correlation coefficient of the i th process parameter when the lag time is seconds; is the average time-lag correlation coefficient of the i th process parameter.

3. The method according to claim 1, wherein, The module comprises the method of any one of claims 1-3.

4. A process parameter timing registration system based on timing correlation matrix and improved particle swarm optimization algorithm, characterized in that, The processor is configured to run a program, wherein the program performs the process parameter time sequence registration method based on the time sequence correlation matrix and the improved particle swarm optimization algorithm of any one of claims 1-3 when running.

5. A processor, comprising: The computer readable storage medium comprises a stored program, wherein the program controls the device where the computer readable storage medium is located to perform the process parameter time sequence registration method based on the time sequence correlation matrix and the improved particle swarm optimization algorithm of any one of claims 1-3 when running.

6. A computer-readable storage medium, characterized in that, ​

Images