A method for distributed operation performance evaluation and non-optimal cause tracing of the whole process of finishing rolling

Abstract

The present invention discloses a method for distributed operation performance evaluation and non-optimal cause tracing of the whole process of finishing rolling, including: obtaining historical sample data retained in the whole process of finishing rolling in the past; dividing the whole process of finishing rolling into three subsystems, namely upstream, midstream and downstream, describing the interdependence between process variables and key performance indicators in each subsystem, removing redundant variables, retaining relevant variables, and obtaining the optimal set of variables of each subsystem; constructing an operation performance evaluation model of each subsystem capable of analyzing the nonlinearity and dynamics of the process and training based on the optimal set of variables of each subsystem; evaluating the operation performance of each subsystem, fusing the evaluation results of each subsystem, realizing global operation performance evaluation, and tracing the non-optimal cause when the process is non-optimal. The present invention can effectively monitor and judge the operation performance of the production process in a timely and accurate manner.

Description

Technical Field

[0001] The present invention relates to the technical field of control and monitoring of industrial production processes, and in particular to a method for evaluating distributed operation performance of the entire finishing rolling process and tracing non-optimal causes. Background Art

[0002] The working conditions in process industries are complex and changeable, and the degradation of their operating performance will lead to a decrease in production efficiency and product quality. Process monitoring has attracted great attention from industry and academia. In order to more effectively monitor operating performance, the concept of key performance indicator (KPI) is introduced. KPI can be roughly divided into three categories: engineering KPI (technical performance and product quality), operation and maintenance KPI (operating speed, maintenance time, cost) and economic KPI (total energy consumption or productivity of the factory). The above KPIs are often difficult to measure directly and there is a significant time delay, such as the outlet thickness, flatness, and convexity of the finishing process. By establishing a quantitative relationship between process variables and KPIs, using process variables to effectively evaluate operating performance is an inevitable trend of process monitoring and an urgent need for improvement.

[0003] Taking the entire process of finishing rolling as an example, we can further deepen our understanding of the necessity and importance of evaluating the operating status of industrial processes.

[0004] The finishing process is inherently nonlinear and dynamic due to frequent changes in product specifications and drastic fluctuations in operating conditions. Load changes change the statistical characteristics of process variables, which in turn causes KPIs to change synchronously with process variables, that is, they have a common trend.

[0005] The entire finishing rolling process is a long series of processes consisting of many equipment, which can be divided into several subsystems. Once the operating performance of any subsystem deteriorates, it may affect the quality of the entire production line and the final product, causing production waste and huge economic losses. Therefore, it is necessary to evaluate its operating performance in real time.

[0006] Mutual Information (MI) is a commonly used evaluation index in information theory to describe the mutual dependence between two variables. It considers the linear and nonlinear relationships between variables. The larger the MI value, the closer the relationship between the variables. However, for a variable set X, it is difficult to obtain the maximum correlation with the target variable c due to the complex calculation of the joint probability density function. By considering the redundancy and correlation between variables, the minimum redundancy maximum correlation (mRMR) algorithm can solve this problem.

[0007] Common trend analysis (CTA) is a non-stationary time series model developed on the basis of cointegration analysis. Each component of the cointegrated multivariate non-stationary process can be decomposed into a non-stationary common trend component and a stationary component. Inspired by CTA, Wu et al. proposed an output-relevant common trend analysis (OCTA) method for KPI-related non-stationary process monitoring. In OCTA, the input and output variables are decomposed into non-stationary common trends and stationary residuals, and the model parameters are estimated by solving the optimization problem.

[0008] Traditional process monitoring methods usually focus on distinguishing whether the process is in normal operation or faulty operation. However, in addition to the above rough classification, on-site operators are more concerned about whether the process is running under optimal conditions. Further evaluating the process operation status, identifying non-optimal causes, and providing guidance and suggestions for maintenance decisions are of great practical significance, which is consistent with the purpose of performance-related process monitoring.

[0009] In recent years, the performance evaluation technology of industrial processes has attracted more and more attention. Evaluation methods can be roughly divided into two categories: evaluation methods based on quantitative information and evaluation methods based on qualitative information. Multivariate statistical analysis is one of the most commonly used quantitative information methods. For production processes with single stable operating conditions, domestic and foreign scholars have proposed performance evaluation methods based on principal component analysis (PCA), multi-set principal component analysis (MPCA) and total latent structure projection (T-PLS). In view of the multi-peak distribution characteristics of process data, the Gaussian mixture model regression (GMMR) method can be used to achieve the optimal evaluation of multimodal process performance. The most commonly used evaluation methods based on qualitative information are Bayesian networks, fuzzy theory and rough set theory. In addition, the combination of quantitative and qualitative methods has gradually been applied to performance evaluation.

[0010] The full-process characteristics of process industries are a huge challenge for operational performance evaluation. Process industries are long series processes consisting of multiple manufacturing units and a series of production equipment, with numerous process variables and coupled correlations. Traditional operational performance evaluation methods are directly applied to the entire process, and it is difficult to obtain satisfactory accuracy. The most common method for dealing with full-process characteristics is the distributed method, which usually divides the process into several subsystems based on process knowledge or variable correlation. However, there is currently little research on the full-process operational performance evaluation of industries, and the main reasons are:

[0011] 1) The process is large-scale, the mechanism is complex, and the inherent dynamics and nonlinearity of the system make it difficult to establish an accurate evaluation model;

[0012] 2) Process variables do not directly affect KPIs, but through a series of intermediate process indicators, so it is difficult to directly extract the correlation between process variables and KPIs;

[0013] 3) When the process is in non-optimal operating conditions, it is difficult to quickly track non-optimal variables.

[0014] Based on the above reasons, the existing operating performance evaluation methods are difficult to complete operating performance evaluation and non-optimal cause tracing in the entire finishing rolling process where nonlinearity and dynamics coexist. Summary of the invention

[0015] The present invention provides a method for distributed operation performance evaluation and non-optimal cause tracing of the whole process of finishing rolling, so as to solve the technical problem that it is difficult to complete the operation performance evaluation and non-optimal cause tracing in the whole process of finishing rolling where nonlinearity and dynamics coexist in the prior art.

[0016] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0017] On the one hand, the present invention provides a method for evaluating the distributed operation performance of the entire process of finishing rolling and tracing the causes of non-optimal rolling, and the method comprises:

[0018] Acquire historical sample data retained in the entire process of finishing rolling in the past, and establish a historical sample database; wherein the historical sample data includes multiple process variables and key performance indicators in the entire process of finishing rolling;

[0019] The whole process of finishing rolling is divided into three subsystems: upstream, midstream and downstream. According to the subsystem to which the historical sample data belongs, the interdependence between process variables and key performance indicators is described in each subsystem, redundant variables are removed, relevant variables are retained, and the optimal set of variables corresponding to each subsystem is obtained.

[0020] Constructing the operation performance evaluation model of each subsystem that can analyze the nonlinearity and dynamics of the process; and training the operation performance evaluation model of each subsystem based on the optimal set of variables of each subsystem;

[0021] Based on the operation performance evaluation model of each subsystem, the operation performance of each subsystem is evaluated respectively;

[0022] The operating performance evaluation results of each subsystem are integrated to achieve global operating performance evaluation. When the production process is non-optimal, the causes of the non-optimality are traced to determine the current non-optimal variables.

[0023] Furthermore, the historical sample data is a data group corresponding to various operating performance levels generated when different steel grades and / or steel plates of different plate thicknesses are generated in the previous full process of finishing rolling; wherein the operating performance levels include four different levels: excellent, good, medium, and poor.

[0024] Furthermore, the entire finishing rolling process includes a first stand, a second stand, a third stand, a fourth stand, a fifth stand, a sixth stand and a seventh stand; the process variables include roll gaps, rolling forces, rolling speeds, currents and voltages of the first stand, the second stand, the third stand, the fourth stand, the fifth stand, the sixth stand and the seventh stand, and bending roll forces of the second stand, the third stand, the fourth stand, the fifth stand, the sixth stand and the seventh stand; the key performance indicators include any one or more combinations of slab thickness, slab width, slab convexity, slab flatness and slab temperature.

[0025] Furthermore, the entire finishing rolling process is divided into three subsystems: upstream, midstream and downstream, including:

[0026] The first, second and third stands in the whole finishing rolling process are divided into an upstream subsystem, the fourth and fifth stands are divided into a midstream subsystem, and the sixth and seventh stands are divided into a downstream subsystem.

[0027] Furthermore, according to the subsystem to which the historical sample data belongs, the interdependence between the process variables and the key performance indicators is described in each subsystem, redundant variables are removed, and relevant variables are retained to obtain the optimal set of variables corresponding to each subsystem, including:

[0028] According to the subsystem to which the historical sample data belongs, the process variables and key performance indicators in the historical sample data are divided into different subsystem data sets: X = [X 上 ,X 中 ,X 下 ],Y=[Y 上 ,Y 中 ,Y 下 ];

[0029] Where X represents the process variable data set, X 上 represents a data set consisting of process variables belonging to the upstream subsystem, X 中 represents the data set consisting of process variables belonging to the midstream subsystem, X 下 represents a data set consisting of process variables belonging to the downstream subsystem; Y represents a key performance indicator data set, Y 上 represents a dataset consisting of key performance indicators belonging to the upstream subsystem, Y 中 represents the data set consisting of the key performance indicators of the midstream subsystem, Y 下A data set representing the key performance indicators belonging to the downstream subsystem;

[0030] The maximum correlation and minimum redundancy between process variables and key performance indicators are calculated in each subsystem, and the joint measurement indicators are constructed as follows:

[0031]

[0032]

[0033]

[0034] In the formula, MIC(x i ,x j ) represents the variable x i With x j The maximum information coefficient of

[0035] Filter out X based on the joint metric B The process variables with the greatest correlation with the corresponding key performance indicators and the least redundancy between each other are used to obtain the optimal set of variables corresponding to each subsystem.

[0036] Furthermore, the operation performance evaluation model of each subsystem capable of analyzing process nonlinearity and dynamics is constructed respectively; and the operation performance evaluation model of each subsystem is trained respectively based on the optimal set of variables of each subsystem, including:

[0037] Will Map to high-dimensional feature space and extract high-dimensional features

[0038] use and Y B Training output related common trend analysis OCTA model:

[0039]

[0040] In the formula, S B and T B Indicates that in subsystem B and Y B The non-stationary trend of and Y B The load matrix, and express and Y B The stationary residuals of and Y B The projection matrix, A T and C T represents the transpose of A and C;

[0041] The optimization function of the OCTA model is obtained as follows:

[0042]

[0043] sA T D=C T E=I

[0044] In the formula, f B (A, C, D, E) represents the parameter optimization function of the B subsystem, and I represents the identity matrix;

[0045] Construct key performance indicators of each subsystem SPE is not related to key performance indicators B as follows:

[0046]

[0047]

[0048] In the formula, Uncovers non-stationary common trends between process variables and key performance indicators; SPE B Used to analyze the stable equilibrium relationship of process variables; Represents the score vector of the new sample point, for The transpose of , ∧ represents the covariance matrix of the process variables, represents the preferred variable for the new sample point, express The high-dimensional features of for The transpose of For S B The transpose of K new is the kernel function matrix of the new sample point, and K0 is the initial kernel function matrix.

[0049] Furthermore, the Map to high-dimensional feature space and extract high-dimensional features include:

[0050] The polynomial kernel function and the radial basis kernel function are integrated as follows to obtain the hybrid kernel function:

[0051]

[0052] In the formula, ω∈[0,1] represents the weight coefficient; represents the mixed kernel function of the optimal variables of sample points i and j, represents the preferred variables of samples i and j in the B subsystem, b is the parameter of the polynomial kernel function, and c is the parameter of the RBF kernel function.

[0053] The hybrid kernel function Map to high-dimensional feature space and extract high-dimensional features

[0054] Furthermore, the information fusion of the operation performance evaluation results of each subsystem to achieve global operation performance evaluation includes:

[0055] According to the thickness deviation of the strip steel product, the running performance is divided into four grades: excellent, good, medium and poor, and the serial numbers 1, 2, 3 and 4 are used to represent the four grades respectively;

[0056] Calculate different levels of and SPE B , and use and SPE B Constructing joint indicators

[0057]

[0058] In the formula, and They are and SPE B The threshold value, The threshold of is obtained by kernel density estimation; and Indicates that the level is c and SPE B

[0059] Using Bayesian reasoning to fuse the performance evaluation results of each subsystem, the probability of the performance being level C is calculated as follows:

[0060]

[0061]

[0062]

[0063] Where, L c Indicates that the operating performance is level C. It means that the operating performance is not level C; represents the posterior probability that the performance of subsystem B is level c, represents the likelihood function that the performance of subsystem B is level c, represents the likelihood function that the performance of subsystem B is not level C, represents the prior probability that the performance of subsystem B is level c, represents the prior probability that the performance of subsystem B is not level C, Indicates that the operating performance is level C. Indicates that the operating performance is not C-level, represents the joint index, Indicates the threshold of the joint indicator.

[0064] Compare the probabilities of different performance levels and determine the operating performance level of subsystem B:

[0065]

[0066] In the formula, It represents the posterior probability that the performance of subsystem B is level c.

[0067] Calculate the global evaluation result based on the evaluation results of each subsystem:

[0068]

[0069] In the formula, is the weight coefficient of subsystem B, round() is the rounding function; β B is the weight coefficient of subsystem B, L B Indicates the operating performance level of subsystem B.

[0070] Furthermore, the implementation of global operation performance evaluation also includes:

[0071] Introduce a moving window with width H;

[0072] The highest-frequency evaluation result among H evaluation results is taken as the operating performance level of the current production process. If there is more than one operating performance level with the highest frequency, the evaluation result of the previous moment is kept unchanged.

[0073] Furthermore, when the production process is not optimal, the reasons for the non-optimization are traced, including:

[0074] When the evaluation result is not optimal, construct non-optimal data With optimal data The difference index dif(i) is used to trace the non-optimal variables:

[0075]

[0076] In the formula, K represents the number of samples of excellent data. The larger the difference index dif(i), the greater the probability that the corresponding variable is a non-optimal variable; x B,i represents the i-th variable of the sample point of subsystem B, Represents the i-th variable of each k-th sample point in the superior data.

[0077] On the other hand, the present invention further provides an electronic device, comprising a processor and a memory; wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the above method.

[0078] In yet another aspect, the present invention further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, and the instruction is loaded and executed by a processor to implement the above method.

[0079] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0080] The full-process distributed operation performance evaluation and non-optimal cause tracing method provided by the present invention will provide new ideas and approaches for the full-process operation performance evaluation of finishing rolling based on data and knowledge, and make up for the dynamic nonlinear process operation performance evaluation and non-optimal cause tracing problems that are difficult to solve with traditional operation performance evaluation methods. It aims to explore practical and effective solutions to the key challenges in operation performance evaluation and non-optimal cause tracing in modern process industries, and has important practical application and promotion value.

[0081] Compared with the traditional operating status evaluation method, the evaluation accuracy of the method of the present invention is improved by 10%. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0083] Figure 1 It is a schematic diagram of the execution flow of the method for evaluating the distributed operation performance of the entire process of finishing rolling and tracing the causes of non-optimal rolling provided by an embodiment of the present invention;

[0084] Figure 2 It is a detailed flow chart of a method for evaluating the distributed operation performance of the entire finishing rolling process and tracing the causes of non-optimal operation provided by an embodiment of the present invention;

[0085] Figure 3 It is a schematic diagram of tracing non-optimal causes provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0086] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0087] First embodiment

[0088] In view of the technical problem that it is difficult to complete the operation performance evaluation and non-optimal cause tracing in the whole process of finishing rolling with both nonlinearity and dynamics in the prior art, this embodiment provides a method for distributed operation performance evaluation and non-optimal cause tracing in the whole process of finishing rolling, which can be implemented by an electronic device, which can be a terminal or a server. The execution process of the method is as follows Figure 1 As shown, the following steps are included:

[0089] S1, obtaining historical sample data retained in the entire process of finishing rolling, and establishing a historical sample database; wherein the historical sample data includes multiple process variables and key performance indicators KPI in the entire process of finishing rolling;

[0090] It should be noted that the historical sample data are data sets of various operating performance levels (excellent / good / medium / poor) generated in the past when steel plates of different steel grades and different plate thicknesses were generated. As shown in Table 1, process variables include the roll gap, rolling force, rolling speed, current, voltage of stands 1-7 and the bending roll force of stands 2-7 (stand 1 has no bending roll force), etc.; KPIs include the thickness, width, convexity, flatness and temperature of the slab.

[0091] Table 1 Finishing process variables

[0092] variable type describe unit 1-7 Process variables Roller gap on working side of frame 1-7 mm 8-14 Process variables Rolling force of stand 1-7 MN 15-20 Process variables Roller bending force of frame 2-7 MN 21-27 Process variables Rolling speed of stand 1-7 mm / s 28-34 Process variables Rack 1-7 Current A 35-41 Process variables Rack 1-7 Voltage KW

[0093] S2, divide the whole process of finishing rolling into three subsystems: upstream, midstream and downstream. According to the subsystem to which the historical sample data belongs, describe the interdependence between process variables and key performance indicators in each subsystem, remove redundant variables, retain relevant variables, and obtain the optimal set of variables corresponding to each subsystem;

[0094] It should be noted that, in this embodiment, Figure 2 As shown, the entire finishing process is divided into three subsystems, namely, upstream, midstream and downstream, according to process knowledge, wherein rack 1, rack 2 and rack 3 are divided into the upstream subsystem, rack 4 and rack 5 are divided into the midstream subsystem, and rack 6 and rack 7 are divided into the downstream subsystem. Specifically, in this embodiment, the above S2 obtains the variable optimization set corresponding to each subsystem by replacing the MI in the traditional mRMR with the maximum information coefficient (MIC) to obtain the improved mRMR: wherein MIC has universality, fairness and symmetry, and the improved mRMR has stronger correlation extraction ability and redundancy removal ability, which is helpful for the construction of the subsequent performance evaluation model. Specifically, the implementation process of the above S2 includes:

[0095] S21, according to the subsystem to which the historical sample data belongs, the process variables and KPIs in the historical sample data are divided into different subsystem data sets: X = [X 上 ,X 中 ,X 下 ],Y=[Y 上 ,Y 中 ,Y 下 ];

[0096] Where X represents the process variable data set, X 上 represents a data set consisting of process variables belonging to the upstream subsystem, X 中 represents the data set consisting of process variables belonging to the midstream subsystem, X 下 represents a data set consisting of process variables belonging to the downstream subsystem; Y represents a key performance indicator data set, Y 上 represents a dataset consisting of key performance indicators belonging to the upstream subsystem, Y 中 represents the data set consisting of the key performance indicators of the midstream subsystem, Y 下 A data set representing the key performance indicators belonging to the downstream subsystem;

[0097] S22, calculate the maximum correlation and minimum redundancy between process variables and KPIs in each subsystem, and construct joint measurement indicators as follows:

[0098]

[0099]

[0100]

[0101] In the formula, MIC(x i ,x j ) represents the variable x i With x j The maximum information coefficient of

[0102] S23, filter out X based on the joint metric B The process variables with the greatest correlation with the corresponding key performance indicators and the least redundancy between each other are used to obtain the optimal set of variables corresponding to each subsystem.

[0103] S3, constructing the operation performance evaluation model of each subsystem that can analyze the nonlinearity and dynamics of the process; based on the optimal set of variables of each subsystem, training the operation performance evaluation model of each subsystem;

[0104] Specifically, in this embodiment, the implementation of S3 is as follows: a new hybrid kernel function is formed by combining a polynomial kernel function and a radial basis kernel function (RBF), and the new hybrid kernel function is combined with OCTA to obtain a KOCTA model based on the hybrid kernel function; wherein the new hybrid kernel function has a stronger generalization ability, and therefore, the constructed KOCTA model has the ability to simultaneously handle process dynamics and nonlinearity, thereby improving the accuracy of the evaluation model. Specifically, S3 includes the following steps:

[0105] S31, in order to better optimize the variable set Mapped to the high-dimensional feature space, this embodiment integrates the polynomial kernel function and the RBF kernel function as follows to obtain a hybrid kernel function:

[0106]

[0107] In the formula, ω∈[0,1] represents the weight coefficient; represents the mixed kernel function of the optimal variables of sample points i and j, represents the preferred variables of samples i and j in the B subsystem, b is the parameter of the polynomial kernel function, and c is the parameter of the RBF kernel function.

[0108] S32, through the hybrid kernel function Map to high-dimensional feature space and extract high-dimensional features

[0109] S33, using high-dimensional features and Y B Training output related common trend analysis OCTA model:

[0110]

[0111] In the formula, S B and T B Indicates that in subsystem B and Y B The non-stationary trend of and Y B The load matrix, and express and Y B The stationary residuals of and Y B The projection matrix, A T and C T Represents the transpose of A and C.

[0112] S34, the optimization function of the KOCTA model can be further obtained as follows:

[0113]

[0114] In the formula, f B (A, C, D, E) represents the parameter optimization function of subsystem B, and I represents the identity matrix.

[0115] S35, build KPI related indicators for each subsystem SPE is not related to KPI B as follows:

[0116]

[0117]

[0118] In the formula, Reveals non-stationary common trends between process variables and KPIs; SPE B Used to analyze the stable equilibrium relationship of process variables; Represents the score vector of the new sample point, for The transpose of , ∧ represents the covariance matrix of the process variables, represents the preferred variable for the new sample point, express The high-dimensional features of for The transpose of For S B The transpose of K new is the kernel function matrix of the new sample point, and K0 is the initial kernel function matrix.

[0119] S4, evaluating the operating performance of each subsystem based on the operating performance evaluation model of each subsystem;

[0120] S5, integrates the operating performance evaluation results of each subsystem to achieve global operating performance evaluation. When the production process is non-optimal, trace the non-optimal causes to determine the current non-optimal variables.

[0121] Specifically, in this embodiment, the acquisition of the global evaluation results and the tracing of the non-optimal causes in the above S5 is to use Bayesian Inference (BI) to fuse the evaluations of different subsystems to obtain the global evaluation results. When the process is non-optimal, the difference index is used to determine the non-optimal variables.

[0122] Specifically, the above S5 includes the following steps:

[0123] S51, according to the thickness deviation of the strip steel product (as shown in Table 2), the running performance is divided into four levels: excellent, good, medium and poor, and the serial numbers 1, 2, 3 and 4 are used to represent the four levels respectively;

[0124] Table 2 Thickness deviation and operating status level

[0125] Thickness(mm) excellent good middle Difference 0.80～2.00 ±0.04 ±0.08 ±0.12 ±0.15 2.00～3.00 ±0.05 ±0.10 ±0.14 ±0.18 3.00～4.00 ±0.06 ±0.12 ±0.17 ±0.22

[0126] S52, calculate different levels of and SPE B , and use and SPE B Constructing joint indicators

[0127]

[0128] In the formula, and They are and SPE B The threshold value, The threshold of can be obtained through kernel density estimation; and Indicates that the level is c and SPE B S53, different operating performance levels have different thresholds. In this regard, this embodiment uses Bayesian reasoning to integrate them, and the probability of the operating performance being level C is calculated as follows:

[0129]

[0130]

[0131]

[0132] Where, L c Indicates that the operating performance is level C. It means that the operating performance is not level C;

[0133] represents the posterior probability that the performance of subsystem B is level c, represents the likelihood function that the performance of subsystem B is level c, represents the likelihood function that the performance of subsystem B is not level C, represents the prior probability that the performance of subsystem B is level c, represents the prior probability that the performance of subsystem B is not level C, Indicates that the operating performance is level C. Indicates that the operating performance is not C-level, represents the joint index, Indicates the threshold of the joint indicator.

[0134] S54, compare the probabilities of different performance levels and determine the operating performance level of subsystem B:

[0135]

[0136] In the formula, It represents the posterior probability that the performance of subsystem B is level c.

[0137] S55, calculating the global evaluation result according to the evaluation results of each subsystem:

[0138]

[0139] In the formula, is the weight coefficient of subsystem B, round() is the rounding function; β B is the weight coefficient of subsystem B, L B Indicates the operating performance level of subsystem B.

[0140] It should be noted that in the actual production process, due to the influence of uncertain factors, it is sometimes difficult to determine the operating performance level through the evaluation results at one time. For this reason, this embodiment introduces a moving window with a width of H; the one with the highest frequency of H evaluation results is taken as the operating performance level of the current production process. If there is more than one operating performance level with the highest frequency, the evaluation result at the previous moment is kept unchanged.

[0141] S56, non-optimal reasons are traced as follows Figure 3 When the evaluation result is not optimal, construct non-optimal data With optimal data The difference index dif(i) between them is used to trace the non-optimal variables:

[0142]

[0143] In the formula, K represents the number of samples of excellent data. The larger the difference index dif(i), the greater the probability that the corresponding variable is a non-optimal variable; x B,i represents the i-th variable of the sample point of subsystem B, Represents the i-th variable of each k-th sample point in the superior data.

[0144] In addition, it should be noted that there are four levels of operating performance: excellent, good, medium and poor. Non-excellent data includes data with operating performance of good, medium and poor levels; the optimal data is data with excellent operating performance.

[0145] In summary, the method for distributed operation performance evaluation and non-optimal cause tracing of the whole process of finishing rolling in this embodiment first uses MIC instead of MI to improve the mRMR algorithm to describe the relationship between process variables and KPIs and reduce the redundancy between variables of each subsystem; secondly, a kernel output-relevant common trend analysis (KOCTA) method based on polynomial kernel function and RBF kernel function is proposed to analyze dynamic and nonlinear characteristics; then, a combined index is designed in each subsystem, and BI is used to fuse all statistical information to accurately evaluate the process operation performance from the perspective of the subsystem and the whole process; finally, when the process is in a non-optimal operating condition, the non-optimal cause can be traced by designing a difference index. Based on this, the method of this embodiment provides a new idea and approach for the operation performance evaluation of the whole process of finishing rolling based on data and knowledge, which makes up for the dynamic nonlinear process operation performance evaluation and non-optimal cause tracing problems that are difficult to solve by traditional operation performance evaluation methods, and has important practical application and promotion value.

[0146] Second embodiment

[0147] This embodiment provides an electronic device, which includes a processor and a memory; wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the method of the first embodiment.

[0148] The electronic device may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) and one or more memories, wherein the memory stores at least one instruction, and the instruction is loaded by the processor to execute the above method.

[0149] Third embodiment

[0150] This embodiment provides a computer-readable storage medium, which stores at least one instruction, and the instruction is loaded and executed by a processor to implement the method of the first embodiment. The computer-readable storage medium may be a ROM, a random access memory, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc. The instructions stored therein may be loaded by a processor in a terminal to execute the method.

[0151] In addition, it should be noted that the present invention can be provided as a method, an apparatus or a computer program product. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0152] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0154] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0155] Finally, it should be noted that the above is a preferred embodiment of the present invention. It should be pointed out that although the preferred embodiment of the present invention has been described, for those skilled in the art, once the basic creative concept of the present invention is known, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the attached claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A method for evaluating the distributed operation performance of the entire finishing rolling process and tracing the causes of non-optimal operation, characterized in that: The method for evaluating the distributed operation performance of the entire finishing rolling process and tracing the causes of non-optimal operation includes: Acquire historical sample data retained in the entire process of finishing rolling in the past, and establish a historical sample database; wherein the historical sample data includes multiple process variables and key performance indicators in the entire process of finishing rolling; The whole process of finishing rolling is divided into three subsystems: upstream, midstream and downstream. According to the subsystem to which the historical sample data belongs, the interdependence between process variables and key performance indicators is described in each subsystem, redundant variables are removed, relevant variables are retained, and the optimal set of variables corresponding to each subsystem is obtained. Constructing the operation performance evaluation model of each subsystem that can analyze the nonlinearity and dynamics of the process; and training the operation performance evaluation model of each subsystem based on the optimal set of variables of each subsystem; Based on the operation performance evaluation model of each subsystem, the operation performance of each subsystem is evaluated respectively; The performance evaluation results of each subsystem are integrated to achieve global performance evaluation. When the production process is not optimal, the cause of the non-optimal process is traced to determine the current non-optimal variables. According to the subsystem to which the historical sample data belongs, the interdependence between the process variables and the key performance indicators is described in each subsystem, redundant variables are removed, relevant variables are retained, and the optimal set of variables corresponding to each subsystem is obtained, including: According to the subsystem to which the historical sample data belongs, the process variables and key performance indicators in the historical sample data are divided into different subsystem data sets: X = [X 上 ,X 中 ,X 下 ],Y=[Y 上 ,Y 中 ,Y 下 ]; Where X represents the process variable data set, X 上 represents a data set consisting of process variables belonging to the upstream subsystem, X 中 represents the data set consisting of process variables belonging to the midstream subsystem, X 下 represents a data set consisting of process variables belonging to the downstream subsystem; Y represents a key performance indicator data set, Y 上 represents a data set consisting of key performance indicators belonging to the upstream subsystem, Y 中 represents the data set consisting of the key performance indicators of the midstream subsystem, Y 下 A data set representing the key performance indicators belonging to the downstream subsystem; The maximum correlation and minimum redundancy between process variables and key performance indicators are calculated in each subsystem, and the joint measurement indicators are constructed as follows: In the formula, MIC(x i ,x j ) represents the variable x i With x j The maximum information coefficient of Filter out X based on the joint metric B The process variables with the greatest correlation with the corresponding key performance indicators and the least redundancy between each other are used to obtain the optimal set of variables corresponding to each subsystem. The operation performance evaluation model of each subsystem capable of analyzing the nonlinearity and dynamics of the process is constructed respectively; and the operation performance evaluation model of each subsystem is trained respectively based on the optimal set of variables of each subsystem, including: Will Map to high-dimensional feature space and extract high-dimensional features use and Y B Training output related common trend analysis OCTA model: In the formula, S B and T B Indicates that in subsystem B and Y B The non-stationary trend of and Y B The load matrix, and express and Y B The stationary residuals of and Y B The projection matrix, A T and C T represents the transpose of A and C; The optimization function of the OCTA model is obtained as follows: In the formula, f B (A, C, D, E) represents the parameter optimization function of the B subsystem, and I represents the identity matrix; Construct key performance indicators of each subsystem SPE is not related to key performance indicators B as follows: In the formula, Uncovers non-stationary common trends between process variables and key performance indicators; SPE B Used to analyze the stable equilibrium relationship of process variables; Represents the score vector of the new sample point, for The transpose of , ∧ represents the covariance matrix of the process variables, represents the preferred variable for the new sample point, express The high-dimensional features of for The transpose of For S B The transpose of K new is the kernel function matrix of the new sample point, K0 is the initial kernel function matrix; When the production process is not optimal, the reasons for the non-optimization are traced, including: When the evaluation result is not optimal, construct non-optimal data With optimal data The difference index dif(i) is used to trace the non-optimal variables: In the formula, K represents the number of samples of excellent data. The larger the difference index dif(i), the greater the probability that the corresponding variable is a non-optimal variable; x B,i represents the i-th variable of the sample point of subsystem B, Represents the i-th variable of each k-th sample point in the superior data.

2. The method for evaluating the distributed operation performance of the entire finishing rolling process and tracing the causes of non-optimal operation according to claim 1, characterized in that: The historical sample data is a data set corresponding to various operating performance levels generated when different steel grades and / or steel plates of different plate thicknesses are generated in the previous full process of finishing rolling; wherein the operating performance levels include four different levels: excellent, good, medium, and poor.

3. The method for distributed operation performance evaluation and non-optimal cause tracing of the whole process of finishing rolling according to claim 1, wherein the whole process of finishing rolling includes a first frame, a second frame, a third frame, a fourth frame, a fifth frame, a sixth frame and a seventh frame; characterized in that: The process variables include the roll gap, rolling force, rolling speed, current and voltage of the first frame, the second frame, the third frame, the fourth frame, the fifth frame, the sixth frame and the seventh frame, and the bending roll force of the second frame, the third frame, the fourth frame, the fifth frame, the sixth frame and the seventh frame; the key performance indicators include any one or more combinations of the thickness of the slab, the width of the slab, the convexity of the slab, the flatness of the slab and the temperature of the slab.

4. The method for evaluating the distributed operation performance of the entire finishing rolling process and tracing the causes of non-optimal operation according to claim 3, characterized in that: The entire finishing rolling process is divided into three subsystems: upstream, midstream and downstream, including: The first, second and third stands in the whole finishing rolling process are divided into an upstream subsystem, the fourth and fifth stands are divided into a midstream subsystem, and the sixth and seventh stands are divided into a downstream subsystem.

5. The method for evaluating the distributed operation performance of the entire finishing rolling process and tracing the causes of non-optimal operation according to claim 1, characterized in that: The Map to high-dimensional feature space and extract high-dimensional features include: The polynomial kernel function and the radial basis kernel function are integrated as follows to obtain the hybrid kernel function: In the formula, ω∈[0,1] represents the weight coefficient; represents the mixed kernel function of the optimal variables of sample points i and j, represents the preferred variables of samples i and j in the B subsystem, b is the parameter of the polynomial kernel function, and c is the parameter of the RBF kernel function; The hybrid kernel function Map to high-dimensional feature space and extract high-dimensional features 6. The method for evaluating the distributed operation performance of the entire finishing rolling process and tracing the causes of non-optimal operation according to claim 5, characterized in that: The information fusion of the operation performance evaluation results of each subsystem to achieve global operation performance evaluation includes: According to the thickness deviation of the strip steel product, the running performance is divided into four grades: excellent, good, medium and poor, and the serial numbers 1, 2, 3 and 4 are used to represent the four grades respectively; Calculate different levels of and SPE B , and use and SPE B Constructing joint indicators In the formula, and They are and SPE B The threshold value, The threshold of is obtained by kernel density estimation; and Indicates that the level is c and SPE B ; Using Bayesian reasoning to fuse the performance evaluation results of each subsystem, the probability of the performance being level C is calculated as follows: Where, L c Indicates that the operating performance is level C. It means that the operating performance is not level C; represents the posterior probability that the performance of subsystem B is level c, represents the likelihood function that the performance of subsystem B is level c, represents the likelihood function that the performance of subsystem B is not level C, represents the prior probability that the performance of subsystem B is level c, represents the prior probability that the performance of subsystem B is not level C, Indicates that the operating performance is level C. Indicates that the operating performance is not C-level, represents the joint index, represents the threshold of the joint indicator; Compare the probabilities of different performance levels and determine the operating performance level of subsystem B: In the formula, represents the posterior probability that the performance of subsystem B is level c; Calculate the global evaluation result based on the evaluation results of each subsystem: In the formula, is the weight coefficient of subsystem B, round() is the rounding function; β B is the weight coefficient of subsystem B, L B Indicates the operating performance level of subsystem B.

7. The method for evaluating the distributed operation performance of the entire finishing rolling process and tracing the causes of non-optimal operation according to claim 6, characterized in that: The implementation of the global operation performance evaluation also includes: Introduce a moving window with width H; The highest-frequency evaluation result among H evaluation results is taken as the operating performance level of the current production process. If there is more than one operating performance level with the highest frequency, the evaluation result of the previous moment is kept unchanged.

Images