Industrial process quality-related monitoring methods based on nonlinear predictable feature analysis

Abstract

This invention discloses an industrial process quality-related monitoring method based on nonlinear predictable feature analysis, belonging to the field of industrial process monitoring technology. The method includes: acquiring production process data under normal operating conditions and performing standardization processing; wherein the production process data includes process variable data and quality variable data; using an improved variable projection importance method, dividing the process variable data into a quality-related subspace and a quality-independent subspace; extracting nonlinear predictable features from the quality-related subspace and the quality-independent subspace respectively; constructing monitoring indicators for the quality-related subspace and the quality-independent subspace; and realizing industrial process quality monitoring based on the monitoring indicators of the quality-related subspace and the quality-independent subspace. This invention can solve the problems of difficulty in quantifying nonlinear correlations in complex industrial processes and the computational curse caused by traditional kernel methods when processing high-order dynamic features.

Description

Technical Field

[0001] This invention relates to the field of data-driven process monitoring technology in industrial production processes, and in particular to a method for monitoring the quality of industrial processes based on nonlinear predictable feature analysis. Background Technology

[0002] With the continuous expansion of modern industry, the requirements for equipment reliability, operational safety, and product quality consistency in production processes have reached unprecedented levels. In complex industrial systems, abnormal fluctuations can not only lead to process instability but also directly affect the quality of the final product and the economic benefits of enterprises. Therefore, how to extract effective information from massive amounts of industrial data to ensure the safety and efficiency of production processes has become a focus of attention for both industry and academia. Against this backdrop, the role of quality-related process monitoring technologies in the industrial field is becoming increasingly prominent.

[0003] In traditional multivariate statistical process control methods, partial least squares (PLS), as a typical data-driven technique, is often used to extract latent features related to quality variables and has shown great potential in practical industrial applications. To further optimize the residual subspace, extended methods such as concurrent PLS have been proposed to separate output-related and input-related information. However, modern industrial processes often contain massive amounts of measurement variables, and the process variables and final product quality variables are not only strongly coupled but also exhibit extremely complex nonlinear and dynamic characteristics. Faced with such high-dimensional, strongly nonlinear, and complex dynamic industrial data, quality monitoring methods relying solely on traditional linear and static models are insufficient to meet the refined and high-quality control requirements of modern industry. Therefore, there is an urgent need to develop quality-related monitoring technologies capable of handling complex industrial processes.

[0004] To improve the performance of quality-related process monitoring, existing technologies have conducted extensive research on nonlinear processing and dynamic characteristics. Targeting the significant nonlinear characteristics of industrial processes, numerous studies have developed nonlinear monitoring models based on kernel principal component analysis and kernel partial least squares. However, directly using all the massive amounts of collected variables to construct a single nonlinear model often introduces a large amount of irrelevant background noise, severely degrading monitoring performance. Therefore, methods based on distributed architectures and utilizing indicators such as mutual information for variable partitioning have gained increasing attention. This architecture aims to divide process variables into different subspaces. While these technologies have improved performance to some extent, existing methods often fail to fully consider the joint nonlinear effects of multiple variables on quality. For example, traditional feature selection methods such as the importance of projected variables are based solely on linear score variance calculations, making it difficult to handle the nonlinear characteristics of the data. Distributed partitioning based on principal component contribution rates or simple mutual information typically only focuses on the relationships within process variables or is limited to measuring the pairwise relationships between a single process variable and quality variables, failing to comprehensively and quantitatively assess the joint nonlinear impact of multiple process variables on product quality. This leads to insufficient nonlinear decoupling during the initial calculation of quality-related variables, easily introducing irrelevant noise or omitting key variables.

[0005] On the other hand, existing quality-related monitoring methods face severe computational bottlenecks when dealing with dynamic characteristics of processes. Modern industrial process data commonly exhibits complex, high-order dynamic characteristics caused by multi-system coupling and closed-loop control. Most current nonlinear quality monitoring models focus on static statistical characteristic analysis, neglecting dynamic time dependencies. To compensate for this deficiency, researchers have extended static models to dynamic methods, such as dynamic principal component analysis, slow feature analysis, and predictable feature analysis (PFA). PFA, a statistical learning method that extracts dynamic features through a latent variable autoregressive model, possesses a unique autoregressive structure that allows it to fit latent variables using multi-step historical information, effectively capturing complex temporal relationships in industrial processes and demonstrating significant advantages in dynamic feature extraction. However, these dynamic methods still have inherent limitations when solving complex industrial problems: most of these dynamic methods do not fundamentally escape the linear assumption. Attempting to extend them to nonlinear versions by introducing traditional implicit kernel methods is problematic because dynamic methods typically require concatenating multi-step historical data to construct a high-dimensional augmented matrix. The computational complexity and memory consumption of such a large implicit kernel data matrix increase exponentially due to the need to calculate and store this matrix. This makes the model computation expensive and lengthy, unable to meet the needs of real-time online monitoring and rapid response under high-frequency sampling in industrial sites.

[0006] In summary, existing quality-related monitoring technologies mainly face the following two major technical challenges: First, there is often a strong nonlinear relationship between process variables and quality variables, which traditional methods find difficult to quantify accurately, resulting in poor quality-related segmentation and feature extraction of variables, and easily causing the loss of key quality information. Secondly, most existing data-driven dynamic methods have not completely escaped the linear assumption. The combination of traditional implicit kernel methods and high-order dynamic models can also lead to extremely serious computational disasters, making it difficult to meet the real-time monitoring needs of industrial sites. Summary of the Invention

[0007] This invention provides an industrial process quality-related monitoring method based on nonlinear predictable feature analysis to solve the technical problems of existing technologies that easily lead to the loss of key quality information and cause extremely serious computational disasters, making it difficult to meet the real-time monitoring needs of industrial sites.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, the present invention provides an industrial process quality-related monitoring method based on nonlinear predictable feature analysis, the industrial process quality-related monitoring method based on nonlinear predictable feature analysis includes: Acquire production process data under normal operating conditions and standardize the acquired production process data; the production process data includes process variable data and quality variable data; Based on standardized production process data, an improved variable projection importance method is used to divide the process variable data into quality-related subspaces and quality-independent subspaces. The improved variable projection importance method introduces mutual information into the variable projection importance method to describe the relationship between process variable data and quality variable data. Nonlinear predictable features are extracted from the quality-related subspace and the quality-independent subspace, respectively; and monitoring indicators are constructed for the quality-related subspace and the quality-independent subspace based on the extracted nonlinear predictable features. Industrial process quality monitoring is achieved based on monitoring indicators in quality-related and quality-independent subspaces.

[0009] Furthermore, based on the standardized production process data, an improved variable projection importance method is used to divide the process variable data into a quality-related subspace and a quality-independent subspace, including: The Partial Least Squares (PLS) algorithm is applied to the standardized production process data to extract the preceding data from the process variable data. The most representative latent variables and their corresponding projected weights; For each extracted latent variable, calculate the sum of squares of its mutual information with all quality variables; The improved projective importance of each latent variable is calculated by replacing the variance explained by the sum of squared mutual information between the latent variable and all quality variables. IVIP ; By all IVIP Variables with values ​​greater than a preset importance threshold constitute a quality-related subspace; consisting of all IVIP Variables whose values ​​are not greater than a preset importance threshold constitute a quality-independent subspace.

[0010] Furthermore, the formula for calculating the sum of squares of mutual information between the latent variable and all quality variables is as follows: ; in, For the first i One hidden variable The sum of squares of mutual information with all quality variables; p The dimension of the quality variable data; Indicates the first i One hidden variable With the j quality variables Mutual information between them.

[0011] Furthermore, the improved formula for calculating the importance of variable projection is as follows: ; in, For the first Projected importance values ​​of each process variable; m The dimension of the process variable data; For the first j The process variable for the first... i The projected weights of the hidden variables; For the first i The projected weight vectors corresponding to the hidden variables.

[0012] Furthermore, the extraction process of nonlinear predictable features includes: Orthogonal random Fourier feature mapping is used to map the data in the subspace from which nonlinear predictable features to be extracted to a high-dimensional nonlinear space; The Predictable Feature Analysis (PFA) algorithm is performed in a high-dimensional nonlinear space to obtain nonlinear predictable features.

[0013] Furthermore, the step of mapping the data within the subspace from which the nonlinear predictable features to be extracted to a high-dimensional nonlinear space using orthogonal random Fourier feature mapping includes: Construct an initial random Gaussian matrix whose elements are independent and identically distributed according to a standard normal distribution. and the matrix Perform QR decomposition to obtain an orthogonal identity matrix. ; Introducing a diagonal scaling matrix Its diagonal elements are sampled independently from The distribution is then used to rescale the basis vectors using the kernel width parameter to obtain the frequency sub-block matrix. : ; in, c For kernel width parameters; To map the data to the target high-dimensional feature space, repeat the above process of constructing the Gaussian matrix, QR decomposition, and scaling reconstruction; until the desired result is generated. A frequency sub-block matrix, wherein... The target mapping dimension is indicated by [], which represents rounding up. Finally, these frequency sub-block matrices are concatenated row-wise and the first row is truncated. The final frequency matrix is ​​obtained by rowing. ; use Perform an explicit orthogonal nonlinear projection transformation on the original process data; combined with Uniformly distributed random bias vector The input data is transformed into a high-dimensional feature representation, as expressed by the formula: ; in, This is the mapped high-dimensional nonlinear feature space data; D The target is mapped to a dimension.

[0014] Furthermore, monitoring indicators for quality-related and quality-independent subspaces are constructed, including: For the subspace of the monitoring indicators to be constructed, among all its nonlinear predictable features, several features corresponding to the smallest eigenvalues ​​are selected and defined as the main predictive features. The remaining features are defined as residual features. ; build , The control limits for each statistic were determined using the kernel density estimation method.

[0015] Furthermore, the monitoring indicators based on quality-related and quality-independent subspaces for industrial process quality monitoring include: Based on monitoring indicators in the quality-related subspace, local monitoring is achieved in the quality-related subspace; Based on monitoring indicators in a mass-independent subspace, local monitoring is achieved in the mass-independent subspace; By using Bayesian inference to fuse the monitoring results of each subspace, global industrial process quality monitoring can be achieved.

[0016] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0017] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the above-described method.

[0018] The beneficial effects of the technical solution provided by this invention include at least the following: 1. Introduction of an improved variable projection importance method: This method overcomes the limitation of traditional methods that can only capture linear variance. By leveraging mutual information, it measures arbitrary correlations between variables from an information theory perspective, comprehensively and quantitatively assessing the joint nonlinear impact of multiple process variables on product quality. It accurately divides the product into quality-related subspaces and independent subspaces, effectively removing a large amount of irrelevant noise and amplifying minute fault signals.

[0019] 2. Independent Monitoring and Traceability in Two Spaces: Anomalies in variables within the quality-related space often directly lead to product degradation, while the independent space reflects changes in local equipment. Establishing separate monitoring models significantly improves the accuracy of process monitoring and enhances fault tracing capabilities.

[0020] 3. Deep Integration of Orthogonal Stochastic Mapping and Predictable Feature Analysis for Extracting Nonlinear Dynamic Features: This invention deeply integrates Orthogonal Stochastic Fourier (ORF) mapping with Predictable Feature Analysis (PFA). ORF mapping imposes strict geometric orthogonality constraints through orthogonal decomposition and scaling reconstruction, explicitly projecting data into a high-dimensional nonlinear space. This overcomes the "curse of computation" caused by traditional implicit kernel methods when processing high-order dynamic models, completely eliminating the exponential computational burden. Meanwhile, PFA, within this high-dimensional orthogonal feature space, fits and maximizes the predictability of latent variables through a multi-order autoregressive model, effectively capturing the complex nonlinear time-series dynamic information of industrial processes.

[0021] 4. Achieving global decision-making through Bayesian fusion strategy: After constructing local statistics for the two types of subspaces, these statistics are fused into a global comprehensive decision index through Bayesian posterior probability inference. This mechanism of separating modeling and decision fusion enhances the comprehensive identification capability of complex failure modes. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the framework of the industrial process quality-related monitoring method based on nonlinear predictable feature analysis provided in the embodiments of the present invention; Figure 2 This is a schematic diagram illustrating the principle of the PFA method; Figure 3 This is a schematic diagram of the implementation process of the industrial process quality-related monitoring method based on nonlinear predictable feature analysis provided in the embodiments of the present invention; Figure 4 This is a system block diagram of the electronic device provided in the embodiments of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0025] First, it should be noted that in the embodiments of the present invention, the words "exemplarily," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplarily" is intended to present the concept in a specific manner. Furthermore, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either one or the other.

[0026] First Embodiment

[0027] To address the difficulties in quantifying nonlinear correlations in complex industrial processes and the "curse of computation" that traditional kernel methods cause when handling high-order dynamic features, this embodiment provides a method for monitoring industrial process quality correlations based on nonlinear predictable feature analysis. This method can be implemented using electronic devices, such as terminals or servers. The framework of this method is as follows: Figure 1 As shown, firstly, an improved variable projection importance method is used to capture the joint nonlinear effect of process variables on quality indicators, accurately defining the variable subspace that influences product quality. Then, an orthogonal stochastic Fourier eigenmap method is proposed to construct a lightweight nonlinear PFA model for efficiently extracting nonlinear dynamic features in the process. Finally, Bayesian inference is used to fuse the monitoring results of each subspace, achieving accurate determination of the global operating state.

[0028] Specifically, the execution flow of this method includes the following steps: Step 1. Quality-related and independent subspace partitioning Given production process data under normal operating conditions, the process data is obtained by standardizing it with zero mean and unit variance. and product quality data ,in, For the number of samples, , The sample dimension is given. The PLS algorithm is applied to the standardized data matrix to extract the top-performing data by maximizing the covariance between input and output variables and cross-validating the prediction error of the quality variables. The most representative latent variables are identified to obtain the latent variable matrix of the process data. And a weight matrix reflecting the contribution of the original variables to the latent variables. To measure arbitrary correlations (including linear and nonlinear) between two random variables, a mutual information estimation algorithm is used instead of the traditional variance explained value. For each extracted latent variable... Calculate its relationship with all quality variables The sum of squares of mutual information: (1) in, This represents the mutual information between two variables calculated using the k-nearest neighbor method. This index quantifies the... The total amount of linear and nonlinear information about product quality contained in the latent variables. Finally, combining the total mutual information correlation of the latent variables and their corresponding projected weights, the final IVIP value is calculated: (2) in, For the first Projected importance values ​​of each process variable; m The dimension of the process variable data; For the first j The process variable for the first... i The projected weights of the hidden variables; For the first i The projected weight vectors corresponding to the hidden variables.

[0029] Given a user-defined variable importance threshold This value is usually 1, which reorganizes and divides the global process variables into two independent subspaces: that is, by all... Quality-related subspace composed of key process variables and by all The quality-independent independent subspace composed of process variables .

[0030] Step 2. Nonlinear mapping

[0031] After partitioning the subspace, to extract nonlinear information during the process, this invention introduces and improves the stochastic Fourier feature map technique, proposing an orthogonal stochastic Fourier feature map (ORF) that includes QR orthogonal decomposition and distribution scale reconstruction. The stochastic Fourier feature map is an explicit kernel approximation method based on Bochner's theorem. From the Gaussian distribution... Independent sampling generation random frequency vectors and construct bias This maps the original data to the feature space: (3) The ORF method aims to eliminate random frequency vectors. The resulting directional redundancy and significantly enhanced statistical independence of the mapped variables are addressed by first constructing an element-independent random Gaussian initial matrix that is identically distributed according to a standard normal distribution. and the matrix Perform QR decomposition to obtain an orthogonal identity matrix. Secondly, in order to maintain the radial density characteristics that the Gaussian distribution should have, a diagonal scaling matrix is ​​introduced. Its diagonal elements are sampled independently from The distribution is due to the norm of a standard Gaussian vector following a certain pattern. Distribution. The basis vectors are rescaled using the kernel width parameter to obtain the frequency sub-block matrix. The construction formula is: (4) in, c For kernel width parameters; To map the data to the target high-dimensional feature space, repeat the above process of constructing the Gaussian matrix, QR decomposition, and scaling reconstruction until the desired feature space is generated. A frequency sub-block matrix, wherein... The target mapping dimension is represented by [], and [] indicates rounding up. Finally, these frequency sub-block matrices are concatenated row-wise and the first row is truncated. The final frequency matrix is ​​obtained by rowing. .

[0032] After completing the construction of the frequency matrix, using Perform an explicit orthogonal nonlinear projection transformation on the original process data. Combined with... Uniformly distributed random bias vector The original low-dimensional spatial data is explicitly projected to a high-dimensional nonlinear feature space; wherein, the input data is transformed into a high-dimensional spatial feature representation as follows: (5) in, This is the mapped high-dimensional nonlinear feature space data; D The target is mapped to a dimension.

[0033] The above parameters c , D The method can be determined by cross-validation to achieve the minimum false alarm rate in a normal sample dataset.

[0034] Step 3. Feature Extraction

[0035] Subsequently, in a high-dimensional nonlinear space, a predictable feature analysis (PFA) model is established through whitening processing and a multi-order autoregressive model. The PFA algorithm is executed, and the whitening space projection matrix is ​​solved through decoupling transformation. Then, combined with the data whitening matrix, the projection matrix of the nonlinear high-dimensional data space is derived. The goal is to minimize the prediction error to obtain the nonlinear predictable features. Specifically, given time series observation data... , Let be the time value. The data undergoes nonlinear mapping and whitening; the whitening transformation formula is: (6) in, For high-dimensional nonlinear data after mapping, This is the whitening matrix. Its expression is: ;symbol This represents the time-domain mean calculation. Then, based on the whitened data, a latent variable autoregressive equation is constructed: (7) in, s Preset the time lag order for the model. for The predicted value of the latent variable at time step. For the front j Latent variables at each time step. Whitening spatial projection matrix. With the autoregression coefficient matrix These are the parameters that need to be solved.

[0036] PFA takes minimizing the prediction error of latent variables as its core optimization objective, and the overall optimization objective function is: (8) Among them, matrix and parameter matrix The two systems are mutually coupled and cannot be solved directly. To address this issue, a solution is introduced... The above formula can be rewritten as: (9) in, This represents multi-step information about history. Equation (9) can be transformed into a regular PCA problem, and solving this problem yields the whitening spatial projection matrix. The projection matrix is ​​then obtained. The PFA method is illustrated as follows: Figure 2 .

[0037] By combining ORF and PFA, an ORFPFA model capable of extracting nonlinear dynamic features is established, thus obtaining nonlinear predictable features: (10) Step 4. Construction of monitoring indicators After obtaining the corresponding subspace using the IVIP block partitioning strategy, the ORF and PFA are combined to establish a nonlinear PFA model. Solving this model yields the nonlinear predictable features: (11) Among all nonlinear prediction features, several dominant features corresponding to the smallest eigenvalues ​​are selected and defined as the primary prediction features. The remaining features are defined as residual features. For these two types of feature spaces, construct... , The control limits were determined using the kernel density estimation method. , .

[0038] To maximize the feature extraction capabilities of PFA, this invention designs the following method for determining the number of features and the step size parameter: The determination of the number of main predictive features is crucial, as the magnitude of the eigenvalues ​​in the PFA method reflects the prediction error introduced by the corresponding eigenvectors. Therefore, it can be achieved through... To determine the number of primary predictive features, gradually increase the number. a， When the cumulative percentage reaches the predetermined limit, such as 90%, select this option. a As the number of main predictive features. Among them, b This represents the total number of PFA eigenvalues. These are the PFA eigenvalues ​​sorted in descending order. Step size parameters The determination of the matrix is ​​denoted by the matrix. The sum of the eigenvalues ​​is , Quantifying variables in The correlation between step delay and time delay. Generally speaking, as... The value increases, It gradually decreases. At this point, the elbow method can be used to determine the optimal parameters. .in, and The firstk Time, Number kj The value of the process variable at time 1 after nonlinear mapping.

[0039] Step 5. Process monitoring workflow based on IVIP-ORFPFA

[0040] Based on the methods in steps 1-4, offline training is performed using collected historical data, such as... Figure 3 As shown, the specific steps are as follows: 1) Collect historical data under normal operating conditions. Taking the wastewater treatment process as an example, the process data includes real-time monitored variables such as influent flow rate, influent ammonia nitrogen concentration, dissolved oxygen concentration in each reaction tank, total suspended solids concentration, and internal reflux rate; product quality data includes core indicators reflecting the final water quality treatment effect, such as effluent nitrate nitrogen concentration. Subsequently, the collected data is standardized with zero mean and unit variance to obtain process data for offline training. and product quality data .

[0041] 2) Execute the variable partitioning strategy based on IVIP described in step 1, and divide the entire process variable set into quality-related subspaces according to the set importance threshold. Mass-independent subspace .

[0042] 3) Based on the nonlinear mapping method described in step 2, explicitly map the data from the two subspaces to a high-dimensional orthogonal feature space to obtain their respective feature mapping functions. Gaussian kernel parameters. It is often set to 200 or higher. The size can be determined by cross-validation to achieve the minimum false alarm rate in a normal sample dataset.

[0043] 4) Based on the process monitoring model based on nonlinear predictive analysis described in step 3, the PFA algorithm is independently executed in the mapped high-dimensional space to extract nonlinear predictable features.

[0044] 5) Based on the monitoring index construction method described in step 4, calculate the local monitoring statistics for the quality correlation space and the independence space on the training set, and use the kernel density estimation method to determine the control limits at a given confidence level. , Set the system to a fault state. Compared to normal state The prior probabilities are respectively , . The significance level corresponding to the control limit can be set to 0.99.

[0045] After offline training, online monitoring is performed using real-time industrial process data to obtain online test samples. These samples are then standardized, subspaced, mapped using high-dimensional features, and statistically calculated. A Bayesian inference strategy is used to fuse the local statistics of each subspace into a global comprehensive inference statistic, which is then compared with a judgment threshold for system anomaly alarms and fault tracing. The specific steps are as follows: 1) For a new online sample First, the data is standardized using the mean and standard deviation of the offline training set. Based on the variable partitioning results determined in step 1, it is then split into quality-correlated and quality-independent subspaces. .

[0046] 2) The split subsamples are input into their respective spaces for high-dimensional mapping and feature extraction according to steps 2-4, and the local online monitoring statistics at the current moment are calculated in real time.

[0047] 3) The conditional probability of each local state is calculated using a Bayesian inference strategy. First, the conditional probability of each local state is calculated. x new In any subspace The following are the conditional probabilities of exhibiting normal and fault states: (12) (13) in, The result calculated for the current sample in this subspace Statistic, This defines the control limits for this subspace. Subsequently, combining the prior probabilities set offline, the posterior fault probability of each subspace is calculated using Bayes' theorem: (14) Finally, using the conditional probability of each subspace inducing a fault as a weight, the global Bayesian comprehensive inference statistic is calculated. : (15) Similarly, statistics can be calculated for the residual feature space. .

[0048] 4) Compare the fused comprehensive statistics with the set fault determination threshold. If... or Exceeding the threshold This triggers a system anomaly alarm. Further, by retrospectively observing the exceedance of local statistics in the quality-related and independent spaces, it can be determined whether the fault occurred in the quality space.

[0049] In summary, this embodiment provides an industrial process quality-related monitoring method based on nonlinear predictable feature analysis. By introducing an improved variable projection importance method, it overcomes the limitation of traditional methods that can only capture linear variance. Utilizing mutual information to measure arbitrary correlations between variables from an information theory perspective, it comprehensively and quantitatively assesses the joint nonlinear impact of multiple process variables on product quality, effectively removing a large amount of irrelevant noise and amplifying minute fault signals. By deeply combining orthogonal stochastic Fourier mapping with predictable feature analysis, it overcomes the computational curse caused by traditional implicit kernel methods when processing high-order dynamic models, completely eliminating the exponential computational burden; and effectively captures complex nonlinear time-series dynamic information of industrial processes. By employing a Bayesian fusion strategy to achieve global decision-making, it enhances the comprehensive identification capability of complex fault modes.

[0050] Second Embodiment

[0051] This embodiment provides an electronic device, such as... Figure 4 As shown, the electronic device includes a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. Furthermore, the electronic device may also include a transceiver, the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0052] Below, in conjunction with Figure 4 A detailed introduction to each component of this electronic device is provided below: The processor is the control center of the electronic device. The electronic device may include multiple processors, each of which can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The term "processor" can refer to a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), other general-purpose processors, application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), one or more field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0053] In a specific implementation, as one example, the processor may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 shown are, of course, merely illustrative examples.

[0054] The memory is used to store the software program that executes the solution of the present invention, and the processor controls its execution. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0055] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may be integrated with the processor or exist independently, and may be accessed through the interface circuit of the electronic device ( Figure 4 (Not shown in the image) is coupled to the processor; however, this embodiment of the invention does not impose specific limitations on this.

[0056] The transceiver may include a receiver and a transmitter. Figure 4 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver can be integrated with the processor or exist independently, and can be connected through the interface circuit of the electronic device (…). Figure 4 (Not shown in the image) is coupled to the processor, and this embodiment of the invention does not specifically limit this.

[0057] In addition, it should be noted that, Figure 4 The structure of the electronic device shown is not intended to limit the device. Actual devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Furthermore, the technical effects achieved by this electronic device when performing the method of the first embodiment described above can be referenced to the technical effects described in the first embodiment; therefore, they will not be repeated here.

[0058] Third Embodiment

[0059] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0060] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely or partially hardware embodiment, a completely or partially software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented in software, 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. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any usable medium accessible to a computer or a data storage device such as a server or data center containing one or more sets of usable media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive (SSD).

[0061] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element. Furthermore, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone, where A and B can be singular or plural. Additionally, the character " / " in this text generally indicates an "or" relationship between the preceding and following objects, but it can also indicate an "AND / OR" relationship. Please refer to the context for specific interpretations. "At least one" refers to one or more items, while "more than" refers to two or more items. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can be represented as: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0064] Furthermore, it is understood that in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0065] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0066] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of functional modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, the functional units in the various embodiments of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0067] If the method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments of the present invention have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make several improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for monitoring the quality of industrial processes based on nonlinear predictable feature analysis, characterized in that, The industrial process quality-related monitoring method based on nonlinear predictable feature analysis includes: Acquire production process data under normal operating conditions and standardize the acquired production process data; the production process data includes process variable data and quality variable data; Based on standardized production process data, an improved variable projection importance method is used to divide the process variable data into quality-related subspaces and quality-independent subspaces. The improved variable projection importance method introduces mutual information into the variable projection importance method to describe the relationship between process variable data and quality variable data. Nonlinear predictable features are extracted from the quality-related subspace and the quality-independent subspace, respectively; and monitoring indicators are constructed for the quality-related subspace and the quality-independent subspace based on the extracted nonlinear predictable features. Industrial process quality monitoring is achieved based on monitoring indicators in quality-related and quality-independent subspaces.

2. The industrial process quality-related monitoring method based on nonlinear predictable feature analysis as described in claim 1, characterized in that, Based on the standardized production process data, an improved variable projection importance method is used to divide the process variable data into a quality-related subspace and a quality-independent subspace, including: The Partial Least Squares (PLS) algorithm is applied to the standardized production process data to extract the preceding data from the process variable data. The most representative latent variables and their corresponding projected weights; For each extracted latent variable, calculate the sum of squares of its mutual information with all quality variables; The improved projective importance of each latent variable is calculated by replacing the variance explained by the sum of squared mutual information between the latent variable and all quality variables. IVIP ; By all IVIP Variables with values ​​greater than a preset importance threshold constitute a quality-related subspace; consisting of all IVIP Variables whose values ​​are not greater than a preset importance threshold constitute a quality-independent subspace.

3. The industrial process quality-related monitoring method based on nonlinear predictable feature analysis as described in claim 2, characterized in that, The formula for calculating the sum of squares of mutual information between the latent variable and all quality variables is: ； in, For the first i One hidden variable The sum of squares of mutual information with all quality variables; p The dimension of the quality variable data; Indicates the first i One hidden variable With the j quality variables Mutual information between them.

4. The industrial process quality-related monitoring method based on nonlinear predictable feature analysis as described in claim 3, characterized in that, The improved formula for calculating the importance of variable projection is as follows: ； in, For the first Projected importance values ​​of each process variable; m The dimension of the process variable data; For the first j The process variable for the first... i The projected weights of the hidden variables; For the first i The projected weight vectors corresponding to the hidden variables.

5. The industrial process quality-related monitoring method based on nonlinear predictable feature analysis as described in claim 1, characterized in that, The extraction process of nonlinear predictable features includes: Orthogonal random Fourier feature mapping is used to map the data in the subspace from which nonlinear predictable features to be extracted to a high-dimensional nonlinear space; The Predictable Feature Analysis (PFA) algorithm is performed in a high-dimensional nonlinear space to obtain nonlinear predictable features.

6. The industrial process quality-related monitoring method based on nonlinear predictable feature analysis as described in claim 5, characterized in that, The step of mapping data within a subspace from which nonlinear predictable features to be extracted using orthogonal random Fourier feature mapping to a high-dimensional nonlinear space includes: Construct an initial random Gaussian matrix whose elements are independent and identically distributed according to a standard normal distribution. and the matrix Perform QR decomposition to obtain an orthogonal identity matrix. ; Introducing a diagonal scaling matrix Its diagonal elements are sampled independently from The distribution is then used to rescale the basis vectors using the kernel width parameter to obtain the frequency sub-block matrix. : ； in, c For kernel width parameters; To map the data to the target high-dimensional feature space, repeat the above process of constructing the Gaussian matrix, QR decomposition, and scaling reconstruction; until the desired result is generated. A frequency sub-block matrix, wherein... The target mapping dimension is indicated by [], which represents rounding up. Finally, these frequency sub-block matrices are concatenated row-wise and the first row is truncated. The final frequency matrix is ​​obtained by rowing. ; use Perform an explicit orthogonal nonlinear projection transformation on the original process data; combined with Uniformly distributed random bias vector The input data is transformed into a high-dimensional feature representation, as expressed by the formula: ； in, This is the mapped high-dimensional nonlinear feature space data; D The target is mapped to a dimension.

7. The industrial process quality-related monitoring method based on nonlinear predictable feature analysis as described in claim 1, characterized in that, The monitoring indicators for constructing quality-related and quality-independent subspaces include: For the subspace of the monitoring indicators to be constructed, among all its nonlinear predictable features, several features corresponding to the smallest eigenvalues ​​are selected and defined as the main predictive features. The remaining features are defined as residual features. ; build , The control limits for each statistic were determined using the kernel density estimation method.

8. The industrial process quality-related monitoring method based on nonlinear predictable feature analysis as described in claim 1, characterized in that, The monitoring indicators based on quality-related and quality-independent subspaces are used to achieve industrial process quality monitoring, including: Based on monitoring indicators in the quality-related subspace, local monitoring is achieved in the quality-related subspace; Based on monitoring indicators in a mass-independent subspace, local monitoring is achieved in the mass-independent subspace; By using Bayesian inference to fuse the monitoring results of each subspace, global industrial process quality monitoring can be achieved.

Images