Multi-mode industrial process monitoring method, device, equipment and storage medium

Abstract

The present invention discloses a multi-mode industrial process monitoring method, apparatus, equipment, and storage medium, including: collecting and preprocessing operating data and performing pattern division through clustering; using affinity matrices between and within patterns to calculate the divergence matrices between and within patterns, respectively; constructing a final objective function; iteratively updating the projection matrix, sharing matrix, and error matrix therein to obtain the final projection matrix; calculating the statistics and control limits of each pattern; calculating the statistics of new samples when they arrive, comparing them with the corresponding control limits, and issuing fault information. High-dimensional data is mapped to a low-dimensional space using the projection matrix, preserving the original features; analyzing the industrial process through the local structure and inherent characteristics of the data within each pattern contained in the features, as well as the relationships between different patterns and common information. The method can fully consider the inherent characteristics of each pattern and the correlations between patterns to monitor the industrial process, thereby improving the accuracy of fault monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process monitoring, and in particular to a multi-mode industrial process monitoring method, device, equipment and storage medium. Background Art

[0002] Multimodal industrial processes operate under varying conditions, depending on factors such as environmental changes, material fluctuations, equipment aging, and economic considerations. These processes exhibit multimodal characteristics depending on the operating conditions or the resulting industrial products. To ensure system reliability and process safety, process detection technology has garnered significant attention in industrial applications. Establishing effective process monitoring mechanisms for complex, multimodal industrial processes is crucial for ensuring production safety and product quality.

[0003] As modern industrial processes become increasingly complex, traditional methods, such as model-based approaches, often involve rigorous analytical model development and are cumbersome to implement. Furthermore, with the advancement of sensor technology and the widespread adoption of distributed control systems, large amounts of historical process data can be easily recorded and stored. Consequently, data-driven process monitoring methods are gaining increasing attention. Multivariate statistical process monitoring methods, such as principal component analysis, independent component analysis, and canonical correlation analysis, can be applied to fault detection in industrial processes with a single mode. However, for multimodal industrial processes, it is often difficult to capture the common features between multiple modes, the local features of each mode, and the correlations between multiple modes. This results in insufficient monitoring accuracy for multimodal industrial processes, making it difficult to meet the production safety and product quality requirements of complex industrial processes. Summary of the Invention

[0004] Embodiments of the present invention provide a multi-mode industrial process monitoring method, apparatus, device, and storage medium to address the technical problem of insufficient accuracy in multi-mode industrial process monitoring due to insufficient capture of local features of each mode and correlation information between modes.

[0005] In a first aspect, an embodiment of the present invention provides a multi-mode industrial process monitoring method, comprising:

[0006] S101, collecting operation data of a multi-mode industrial process and preprocessing the operation data to obtain preprocessed operation data;

[0007] S102, performing cluster analysis on the pre-processed operation data, dividing the modes according to the clustering results, and obtaining operation data of each mode;

[0008] S103, constructing an inter-mode affinity matrix and an intra-mode affinity matrix using the operation data of each mode, and calculating an inter-mode scatter matrix and an intra-mode scatter matrix based on the inter-mode affinity matrix and the intra-mode affinity matrix;

[0009] S104, constructing a final objective function based on the inter-mode scatter matrix and the intra-mode scatter matrix, wherein the final objective function reflects the intra-mode information, inter-mode information, and common information of each mode industrial process, and extracting the projection matrix, the shared matrix, and the error matrix using the final objective function;

[0010] S105, iteratively updating the projection matrix, the shared matrix, and the error matrix according to the final objective function to obtain the final projection matrix;

[0011] S106, using the final projection matrix to calculate the statistics and control limits of each mode, when a new sample arrives, by calculating the statistics of the new sample and comparing it with the control limits of its corresponding mode, an industrial process fault message is issued when the statistics exceed the control limits.

[0012] Furthermore, the S104 includes:

[0013] The final objective function is constructed using the inter-mode scatter matrix and the intra-mode scatter matrix;

[0014] The augmented Lagrangian function is introduced to solve the final objective function and the augmented Lagrangian function of the final objective function is obtained;

[0015] The projection matrix, shared matrix and error matrix are extracted using the augmented Lagrangian function of the final objective function.

[0016] Furthermore, the constructing of the objective function using the inter-mode scatter matrix and the intra-mode scatter matrix includes:

[0017] The objective function is constructed by using the inter-mode divergence matrix and the intra-mode divergence matrix and introducing the idea of ​​representation learning.

[0018] Furthermore, the S105 includes:

[0019] According to the final objective function, the projection matrix and error matrix in the final objective function are fixed, and the shared matrix is ​​updated;

[0020] Fix the projection matrix and the updated shared matrix, and update the error matrix;

[0021] Fix the updated shared matrix and the updated error matrix, and update the projection matrix;

[0022] The updated projection matrix is ​​compared with a preset error threshold to determine the final projection matrix.

[0023] Furthermore, the method further comprises:

[0024] Update the Lagrange multipliers and penalty parameters based on the updated projection matrix, the updated shared matrix, and the updated error matrix.

[0025] Furthermore, the S103 includes:

[0026] All sample data in the operation data of each mode are used to form a multi-mode industrial process matrix. The Euclidean distance between each sample in the multi-mode industrial process matrix is ​​used to construct the inter-mode affinity matrix and the intra-mode affinity matrix respectively.

[0027] The inter-mode scatter matrix and the intra-mode scatter matrix are calculated based on the inter-mode affinity matrix and the intra-mode affinity matrix.

[0028] Furthermore, the S106 includes:

[0029] Use the final projection matrix to calculate the Statistics, SPE Statistics and control limits;

[0030] For the new sample obtained, first determine the mode to which the new sample belongs, and use the final projection matrix to calculate the new sample's Statistics and SPE Statistics;

[0031] The new sample Statistics and SPE The statistic is compared with the control limits corresponding to the mode to which the new sample belongs, and an industrial process failure message is issued when the statistic exceeds the control limits.

[0032] In a second aspect, an embodiment of the present invention provides a multi-mode industrial process monitoring device, comprising:

[0033] Data collection module, used to collect and pre-process the operating data of multi-mode industrial processes;

[0034] Data classification module, used to perform pattern classification on pre-processed operation data through cluster analysis;

[0035] A matrix calculation module is used to construct an inter-mode affinity matrix and an intra-mode affinity matrix according to the operation data of each mode and calculate a divergence matrix;

[0036] A final objective function calculation module, used for forming a final objective function according to the inter-mode scatter matrix and the intra-mode scatter matrix;

[0037] Function iteration module, used to iteratively calculate the final projection matrix of the final objective function;

[0038] The fault diagnosis module is used to calculate the statistics and control limits of each mode according to the projection matrix, and to calculate the statistics of the new sample and compare them with the control limits of the corresponding mode to issue industrial process fault information.

[0039] In a third aspect, an embodiment of the present invention provides a computer device, including:

[0040] one or more processors;

[0041] a storage device for storing one or more programs,

[0042] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned multi-mode industrial process monitoring method.

[0043] In a fourth aspect, an embodiment of the present invention provides a storage medium comprising computer-executable instructions, which, when executed by a computer processor, are used to perform the above-mentioned multi-mode industrial process monitoring method.

[0044] The embodiment of the present invention provides a multi-mode industrial process monitoring method, device, equipment and storage medium. The method calculates the inter-mode divergence matrix and the intra-mode divergence matrix through the inter-mode affinity matrix and the intra-mode affinity matrix, and constructs the final objective function; extracts the projection matrix, the sharing matrix and the error matrix from the final objective function, and obtains the final projection matrix through iterative updating; uses the final projection matrix to calculate the Statistics, Statistics and control limits. During the monitoring phase, the final projection matrix is ​​used to calculate the statistics of new samples and compare them with the control limits of their corresponding modes to determine whether a fault exists, thereby enabling online fault monitoring of multi-mode industrial processes. The projection matrix maps high-dimensional data to a low-dimensional space, ensuring the preservation of original features. By analyzing the local structure of the data within each mode contained in the features, the characteristics of each mode itself, the relationships between different modes, and the common information between multiple modes, the industrial process is analyzed to identify whether the data sample of the industrial process is a fault sample. This fully considers the characteristics of each mode and the correlation between modes, resulting in more accurate identification results and improved fault monitoring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are provided to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0046] Figure 1 This is a flow chart of a multi-mode industrial process monitoring method according to the first embodiment of the present invention;

[0047] Figure 2 This is a flow chart of a multi-mode industrial process monitoring method according to the second embodiment of the present invention;

[0048] Figure 3 This is a flow chart of a multi-mode industrial process monitoring method according to the third embodiment of the present invention;

[0049] Figure 4 This is a schematic structural diagram of a multi-mode industrial process monitoring device according to a fourth embodiment of the present invention;

[0050] Figure 5 This is a structural diagram of the computer device described in Example 5 of the present invention. DETAILED DESCRIPTION

[0051] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0052] A multimodal industrial process typically operates under a single core production system, adjusting operating conditions, control strategies, or operational parameters to produce different product specifications or meet varying production needs. For example, within the same processing equipment system (e.g., a catalytic cracking unit) and basic process flow, by adjusting key process parameters (e.g., reaction temperature, catalyst activity, fractionation cut point, etc.), different oil product specifications (e.g., 92# gasoline and 95# gasoline) can be produced without changing the core process route. Each mode in a multimodal industrial process possesses local information within it, and certain interrelated information exists between modes. By capturing this local and interrelated information, the unique characteristics of each mode and their interrelated characteristics can be fully considered, enabling more accurate analysis of industrial process data and more precise monitoring of the process status.

[0053] Example 1

[0054] Figure 1 This is a flow chart of a multi-mode industrial process monitoring method according to a first embodiment of the present invention. This embodiment constructs a function by using local information of each mode and the association between the modes to analyze whether there is a fault in the operating state. The method specifically includes the following steps:

[0055] S101 , collecting operation data of a multi-mode industrial process, and preprocessing the operation data to obtain preprocessed operation data.

[0056] In a multi-mode industrial process, operating data is collected as data samples. The collected sample data is preprocessed using a normalization method, and data in different ranges are converted to a unified scale range to eliminate the scale differences between variables and generate preprocessed operating data to facilitate subsequent calculation and analysis.

[0057] S102 , performing cluster analysis on the pre-processed operation data, dividing the modes according to the clustering results, and obtaining operation data of each mode.

[0058] The preprocessed operating data is clustered using a clustering algorithm. For example, the preprocessed operating data can be processed using a k-means algorithm. The data is divided into operating data of different modes according to the clustering results. The local information of each mode itself and the correlation information between the modes can be analyzed based on the divided data.

[0059] S103 , constructing an inter-mode affinity matrix and an intra-mode affinity matrix respectively using the operation data of each mode, and calculating an inter-mode divergence matrix and an intra-mode divergence matrix according to the inter-mode affinity matrix and the intra-mode affinity matrix.

[0060] The inter-mode affinity matrix is ​​used to describe the relationship between data of different modes and capture the geometric structure relationship of data between modes; the intra-mode affinity matrix is ​​used to describe the relationship between data within a single mode, which can maintain the local structure of the data so that adjacent data points remain adjacent after data dimensionality reduction, thereby capturing the local information of the data within the mode; the inter-mode scatter matrix is ​​used to describe the degree of data separation between different modes; the intra-mode scatter matrix is ​​used to describe the degree of data separation within a single mode. The classified operation data of all mode industrial processes can be expressed as ,in m Indicates the number of variables, k Indicates the number of modes, n represents the number of samples of a single pattern, Indicates the i The affinity matrix between patterns is constructed by using the classification operation data of all patterns. and intra-pattern affinity matrix :

[0061]

[0062]

[0063] in, i Indicates the i The serial number of the sample, j Indicates the j The serial number of the sample, Indicates the isamples, Indicates the j samples, represents the total number of samples, Indicates the k The number of samples for each mode, Indicates the k The collection of all data of a model, represents the weight between two adjacent data samples, Represents a sample The surrounding local scale, Representation sample No. q Nearest neighbor samples, Determination of same.

[0064] Using the inter-mode affinity matrix and the intra-mode affinity matrix, we construct the inter-mode divergence matrix by accumulating the product of affinity and Euclidean distance between all samples. and the intra-mode scatter matrix :

[0065]

[0066]

[0067] in, i Indicates the i The serial number of the sample, j Indicates the j The serial number of the sample, Indicates the i samples, Indicates the j samples, represents the total number of samples, represents the affinity matrix between patterns, represents the intra-pattern affinity matrix, Indicates the transpose of the matrix. Since any two samples are calculated twice during the traversal calculation process, the accumulated result needs to be multiplied by 1 / 2 to eliminate duplication.

[0068] S104, constructing a final objective function based on the inter-mode scatter matrix and the intra-mode scatter matrix, the final objective function reflects the intra-mode information, inter-mode information and common information of each mode industrial process, and using the final objective function to extract the projection matrix, shared matrix and error matrix.

[0069] In order to comprehensively consider the local structure of data within each mode, the characteristics of each mode, the relationship between different modes, and the common information between multiple modes in the multi-mode industrial process, the inter-mode scatter matrix and the intra-mode scatter matrix are used to construct the final objective function:

[0070]

[0071]

[0072] in, represents the projection matrix, which is the mapping of high-dimensional data to low-dimensional data and preserves the data characteristics, and , is the real number space, m Indicates the number of variables, d represents the data dimension after dimensionality reduction, Indicates transposing the projection matrix. represents the equilibrium parameter, represents the trace of the matrix, It means to find the minimum value when satisfying the objective function. , represents the shared matrix, and , n represents the number of samples of a single pattern, Represents the shared matrix C of F The square of the norm, and is the penalty parameter, Indicates the i The error matrix of each mode is , Indicates the i Model error matrix The 2,1 norm of is used to make the error matrix sparse.

[0073] S105 , according to the final objective function, iteratively update the projection matrix, the shared matrix, and the error matrix respectively to obtain the final projection matrix.

[0074] In order to maximize the accuracy of data analysis and fully extract the characteristics of each pattern and the correlation between them, the projection matrix needs to be iteratively updated. For example, the ADMM algorithm (Alternating Direction Method of Multipliers) can be used for iterative updates: first, the projection matrix and error matrix are fixed, and the shared matrix is ​​updated; then, the projection matrix and the updated shared matrix are fixed, and the error matrix is ​​updated; then, the updated shared matrix and the error matrix are fixed again, and the projection matrix is ​​updated; finally, the penalty parameter is updated and compared with the preset stopping condition to determine whether the condition is met and stop updating or repeat the above iterative process until the stopping condition is met, and output the final projection matrix.

[0075] S106, using the final projection matrix to calculate the statistics and control limits of each mode, when a new sample arrives, by calculating the statistics of the new sample and comparing it with the control limits of its corresponding mode, an industrial process fault message is issued when the statistics exceed the control limits.

[0076] According to the final projection matrix, construct each mode Statistics and SPE The statistics monitor the principal component space and the residual space separately. A statistic is a statistic used for multivariate data sets to determine whether a given observation deviates significantly from the central position of other observations. SPE Statistics are a measure used to monitor prediction errors in multivariable systems. Their physical meaning is to monitor whether there are anomalies by comparing the differences between observed data and model predicted data under normal operating conditions.

[0077] It should be noted that the control limit represents the critical value between the normal operating state and the fault state in the industrial process, and is mainly used to monitor whether a fault occurs. During the model training process, the control limit is determined by the preset confidence level using the operating data under normal operating conditions. For example, when the confidence level is 99%, the control limit is the critical value between the normal operating state and the fault state. The statistic is 30 (calculated from the training data 99% of the statistics are below 30), 30 can be determined as The control limits for the statistic.

[0078] When monitoring industrial processes online, real-time monitoring of the fault status of the industrial process is performed: when a new sample When it arrives, it is first preprocessed and the distance between the sample and the center of each pattern is calculated. The pattern with the shortest distance is determined as the pattern to which the sample belongs. Then the final projection matrix obtained in the training phase is used to calculate Statistics and Statistics are collected and compared with the control limits of the corresponding mode. Whether it exceeds the control limits is determined to determine whether it is a fault sample, and fault information is sent to the staff in real time.

[0079] The embodiment of the present invention provides a multi-mode industrial process monitoring method, device, equipment and storage medium. The method calculates the inter-mode divergence matrix and the intra-mode divergence matrix through the inter-mode affinity matrix and the intra-mode affinity matrix, and constructs the final objective function; extracts the projection matrix, the sharing matrix and the error matrix from the final objective function, and obtains the final projection matrix through iterative updating; uses the final projection matrix to calculate the Statistics, Statistics and control limits. During the monitoring phase, the final projection matrix is ​​used to calculate the statistics of new samples and compare them with the control limits of their corresponding modes to determine whether a fault exists, thereby enabling online fault monitoring of multi-mode industrial processes. The projection matrix maps high-dimensional data to a low-dimensional space, ensuring the preservation of original features. By analyzing the local structure of the data within each mode contained in the features, the characteristics of each mode itself, the relationships between different modes, and the common information between multiple modes, the industrial process is analyzed to identify whether the data sample of the industrial process is a fault sample. This fully considers the characteristics of each mode and the correlation between modes, resulting in more accurate identification results and improved fault monitoring accuracy.

[0080] Example 2

[0081] Figure 2 This is a flow chart of a multi-mode industrial process monitoring method according to the second embodiment of the present invention. This embodiment is optimized based on the above embodiment. In this embodiment, S104 is specifically optimized as follows:

[0082] The final objective function is constructed using the inter-mode scatter matrix and the intra-mode scatter matrix;

[0083] The augmented Lagrangian function is introduced to solve the final objective function and the augmented Lagrangian function of the final objective function is obtained;

[0084] The projection matrix, shared matrix and error matrix are extracted using the augmented Lagrangian function of the final objective function.

[0085] Accordingly, the multi-mode industrial process monitoring method provided in this embodiment specifically includes:

[0086] S201 , collecting operation data of a multi-mode industrial process, and preprocessing the operation data to obtain preprocessed operation data.

[0087] S202 , performing cluster analysis on the pre-processed operation data, dividing the modes according to the clustering results, and obtaining operation data of each mode.

[0088] S203 , constructing an inter-mode affinity matrix and an intra-mode affinity matrix respectively using the operation data of each mode, and calculating an inter-mode divergence matrix and an intra-mode divergence matrix according to the inter-mode affinity matrix and the intra-mode affinity matrix.

[0089] S204, constructing a final objective function using the inter-mode scatter matrix and the intra-mode scatter matrix.

[0090] The inter-mode scatter matrix can describe the degree of separation of data between different modes, and the intra-mode scatter matrix can describe the degree of separation of data within a mode. When constructing the final objective function, the inter-mode scatter is as large as possible, and the intra-mode scatter is as small as possible, so that different modes are more separable in low-dimensional space while maintaining the data structure of each mode. The constructed final objective function can reflect the maintenance of local data structure, the inherent characteristics of each mode, and the relationship between modes:

[0091]

[0092] in, P represents the projection matrix, which is the mapping of high-dimensional data to low-dimensional data and preserves the data characteristics, and , is the real number space, m Indicates the number of variables, d represents the data dimension after dimensionality reduction, T represents the transpose of the matrix, Indicates transposing the projection matrix. represents the inter-mode divergence matrix, represents the intra-mode scatter matrix.

[0093] Specifically, the method of constructing the final objective function using the inter-mode scatter matrix and the intra-mode scatter matrix includes:

[0094] The final objective function is constructed by using the inter-mode divergence matrix and the intra-mode divergence matrix and introducing the idea of ​​representation learning.

[0095] By introducing the idea of ​​representation learning into the constructed objective function, a new final objective function is formed:

[0096]

[0097]

[0098] in, is the constraint condition of the final objective function, used to extract the shared matrix And make the projection matrix and the shared matrix affect each other. Used to capture information within and between modes; Aims to use the learned shared matrix C To capture common information among multiple modes; It is to make The columns of are zero, making the error matrix sparse, reducing the sensitivity to outliers, and improving the anti-interference and robustness of the model.

[0099] and By introducing representation learning, the model automatically learns the common features between different modes in the data, reducing manual intervention and improving model performance. The new final objective function comprehensively considers the local structure of multimodal industrial process data, the characteristics of each mode, the relationship between two modes, and the common information among multiple modes.

[0100] S205 , introducing an augmented Lagrangian function to solve the final objective function, and obtaining an augmented Lagrangian function of the final objective function.

[0101] Since the solution process of the final objective function is relatively complicated, in order to simplify the solution process, the augmented Lagrangian function is used to optimize the final objective function:

[0102]

[0103] in, represents the augmented Lagrangian function of the final objective function, Indicates the i The Lagrange multiplier of the mode, and , k Indicates the number of modes, i Indicates the i The serial number of the mode, Indicates the i A low-dimensional representation of the pattern data, Represents the first i A low-dimensional representation of the pattern data, represents the error matrix, represents the Lagrangian term of the linear constraint, represents the penalty parameter, F express F norm, represents the augmented Lagrangian penalty term.

[0104] By simplifying and combining the above formulas, the augmented Lagrangian function of the final objective function can be reconstructed as:

[0105]

[0106] in, represents the reconstruction error constraint.

[0107] S206 , extracting the projection matrix, the shared matrix, and the error matrix using the augmented Lagrangian function of the final objective function.

[0108] In the augmented Lagrangian function of the reconstructed final objective function, the projection matrix is P , used to represent the mapping of high-dimensional data to low-dimensional space, extracting features from high-dimensional data and retaining the original features in low-dimensional data, which is convenient for optimization and iterative operations; the shared matrix is C , used to represent the common features between different modes. Since the equipment, raw materials, processes, etc. between different modes are the same or similar, there will be some common points between different modes; the error matrix is , which represents the error when reconstructing high-dimensional data in a low-dimensional space. This error can be used to optimize the projection matrix so that the projected data can more comprehensively describe the characteristics of the original data, improving the accuracy of the model. The ADMM algorithm can be used to iteratively update the final projection matrix.

[0109] S207 , according to the final objective function, iteratively update the projection matrix, the shared matrix, and the error matrix respectively to obtain the final projection matrix.

[0110] S208, using the final projection matrix to calculate the statistics and control limits of each mode, when a new sample arrives, by calculating the statistics of the new sample and comparing it with the control limits of its corresponding mode, an industrial process fault message is issued when the statistics exceed the control limits.

[0111] This embodiment introduces the idea of ​​representation learning, constructs a new final objective function, and uses the augmented Lagrangian function to solve the optimization problem of the final objective function, so that the model can automatically learn useful feature representations in the data, reduce manual intervention, improve the performance of the model, and at the same time simplify the solution process of the final objective function, reduce computational complexity, and improve the computational efficiency of the model.

[0112] In an optional implementation of this embodiment, S103 includes:

[0113] All sample data in the operation data of each mode are used to form a multi-mode industrial process matrix. The Euclidean distance between each sample in the multi-mode industrial process matrix is ​​used to construct the inter-mode affinity matrix and the intra-mode affinity matrix respectively.

[0114] When constructing the inter-mode affinity matrix and the intra-mode affinity matrix, the geometric structures between modes and within modes are captured respectively, that is, the Euclidean distance between samples in the high-dimensional data space composed of samples is used to construct the inter-mode affinity matrix and the intra-mode affinity matrix respectively.

[0115] The inter-mode scatter matrix and the intra-mode scatter matrix are calculated based on the inter-mode affinity matrix and the intra-mode affinity matrix.

[0116] By accumulating the product of affinity and Euclidean distance between all samples, we construct the inter-pattern scatter matrix and the intra-pattern scatter matrix. The inter-pattern affinity matrix reflects the association between patterns, while the intra-pattern affinity matrix reflects the local characteristics of the data within a single pattern, which is used to retain the original characteristics after data dimensionality reduction. The inter-pattern scatter matrix reflects the differences between patterns, while the intra-pattern scatter matrix reflects the degree of separation of data within a pattern. When using the scatter matrix to construct the final objective function, the inter-pattern scatter is as large as possible and the intra-pattern scatter is as small as possible, thereby making different patterns more separable in low-dimensional space while maintaining the local structural characteristics of the data within each pattern.

[0117] Example 3

[0118] Figure 3 This is a flow chart of a multi-mode industrial process monitoring method according to the third embodiment of the present invention. This embodiment is optimized based on the above embodiment. In this embodiment, S105 is specifically optimized as follows:

[0119] According to the final objective function, the projection matrix and error matrix in the final objective function are fixed, and the shared matrix is ​​updated;

[0120] Fix the projection matrix and the updated shared matrix, and update the error matrix;

[0121] Fix the updated shared matrix and the updated error matrix, and update the projection matrix;

[0122] The updated projection matrix is ​​compared with a preset error threshold to determine the final projection matrix.

[0123] Accordingly, the multi-mode industrial process monitoring method provided in this embodiment specifically includes:

[0124] S301 , collecting operation data of a multi-mode industrial process, and preprocessing the operation data to obtain preprocessed operation data.

[0125] S302 , performing cluster analysis on the pre-processed operation data, dividing the modes according to the clustering results, and obtaining operation data of each mode.

[0126] S303 , constructing an inter-mode affinity matrix and an intra-mode affinity matrix respectively using the operation data of each mode, and calculating an inter-mode divergence matrix and an intra-mode divergence matrix according to the inter-mode affinity matrix and the intra-mode affinity matrix.

[0127] S304, constructing a final objective function based on the inter-mode scatter matrix and the intra-mode scatter matrix, wherein the final objective function reflects the intra-mode information, inter-mode information and common information of each mode industrial process, and using the final objective function to extract the projection matrix, shared matrix and error matrix.

[0128] S305 , according to the final objective function, fixing the projection matrix and the error matrix in the final objective function, and updating the shared matrix.

[0129] By fixing the projection matrix P and the error matrix , solve the shared matrix update function obtained by the final objective function transformation to update the shared matrix C To update:

[0130]

[0131] in, represents the penalty parameter, F express F norm, represents the penalty parameter, i Indicates the i mode, k Indicates the number of modes, represents the transpose of the projection matrix, Indicates the i The data of the pattern, Indicates the i A low-dimensional representation of the pattern data, Represents the first i A low-dimensional representation of the pattern data, Indicates the i The Lagrange multiplier of the mode, represents the reconstruction error constraint in the final objective function.

[0132] By taking the shared matrix C The partial derivatives of are set to zero: , and get the updated shared matrix :

[0133]

[0134]

[0135] in, represents the identity matrix, represents the projection matrix, Indicates the i Transpose of the pattern data.

[0136] S306: Fix the projection matrix and the updated shared matrix, and update the error matrix.

[0137] By fixing the projection matrix P and the updated shared matrix , solve the error matrix update function obtained by the final objective function transformation to update the error matrix To update:

[0138]

[0139] in, represents the penalty parameter, Indicates the final objective function i The reconstruction error constraints of the modes. The partial derivative of is:

[0140]

[0141] in, represents a diagonal matrix derived from the error matrix.

[0142] By taking the error matrix The partial derivatives of are set to zero to obtain the updated error matrix :

[0143]

[0144] S307: Fix the updated shared matrix and the updated error matrix, and update the projection matrix.

[0145] By fixing the updated shared matrix and the updated error matrix , solve the projection matrix update function obtained by the final objective function to update the projection matrix P To update:

[0146]

[0147] in, represents the inter-mode divergence matrix, represents the intra-mode scatter matrix, represents the equilibrium parameter, Represented in the projection matrix The trace of the difference between the intra-modal divergence of the data and the adjusted inter-modal divergence, represents the reconstruction error constraint in the final objective function.

[0148] By taking the projection matrix in the above formula P The partial derivatives of are set to zero: , get the updated projection matrix :

[0149]

[0150] An optional implementation of this embodiment is that the method further includes:

[0151] Update the Lagrange multipliers and penalty parameters based on the updated projection matrix, the updated shared matrix, and the updated error matrix.

[0152] After updating the projection matrix, shared matrix, and error matrix once respectively, the Lagrange multiplier and penalty parameter also need to be updated:

[0153]

[0154] in, t Indicates the t Step update, Indicates in t The first update obtained i The Lagrange multiplier of the mode, Indicates in t +1 step update i The Lagrange multiplier of the mode, Denotes the constrained residual. t The Lagrange multiplier, projection matrix, shared matrix, and error matrix obtained during the step update are used to calculate the t +1 step update Lagrange multiplier. Similarly, t The penalty parameter obtained during the step update Multiply by the step size Then the maximum value of the penalty parameter is set Compare the two and select the minimum value as the t +1 step update penalty parameter .

[0155] The above iterative update process is the ADMM algorithm. After updating the projection matrix, shared matrix, error matrix, Lagrange multiplier and penalty parameter once, a round of update is completed. Subsequently, the decision on whether to perform the next round of update is made by comparing with the threshold. The iteration is repeated until the comparison with the threshold satisfies the condition and the update is stopped.

[0156] S308 : Determine a final projection matrix by comparing the updated projection matrix with a preset error threshold.

[0157] The updated projection matrix is ​​used to construct a projection error function, and the projection error function is compared with the preset error threshold:

[0158]

[0159] When the projection error function is less than the error threshold When , it can be considered that the stopping condition is met, the iterative update process is stopped, and the latest projection matrix is ​​output as the final projection matrix. For example, the convergence condition of the iterative update can be determined based on past experience as The final projection matrix can be used to calculate the statistics and control limits of each mode. When performing online monitoring, the statistics of the new sample can be compared with the control limits of its corresponding mode to analyze the operating status of the industrial process.

[0160] S309, using the final projection matrix to calculate the statistics and control limits of each mode. When a new sample arrives, the statistics of the new sample are calculated and compared with the control limits of its corresponding mode. When the statistics exceed the control limits, an industrial process fault message is issued.

[0161] This embodiment uses the projection matrix, shared matrix, and error matrix extracted from the final objective function, and adopts the ADMM algorithm to iteratively update the projection matrix, shared matrix, and error matrix respectively until the error meets the requirements, and outputs the latest projection matrix as the final projection matrix. The final projection matrix can be used to calculate the statistics and control limits of each mode. When performing online monitoring, the statistics of the new sample are compared with the control limits of its corresponding mode to analyze the operating status of the industrial process. The iterative calculation process is to continuously reduce the error of mapping high-dimensional data to low-dimensional space until it approaches the actual situation. At the same time, it can fully consider the local characteristics of each mode itself and the common information between them. During the iterative update process, the projection matrix is ​​continuously calibrated to obtain a more accurate projection matrix, which can better retain the original characteristics, thereby more accurately deriving statistics, making the monitoring results of the industrial mode more accurate.

[0162] Optionally, the S106 includes:

[0163] Use the final projection matrix to calculate the Statistics, SPE Statistics and control limits;

[0164] Using the final projection matrix, we project the data sample into the principal component space and calculate the Mahalanobis distance between its principal component score and the center of the principal component space to construct Statistics monitor the principal component space; by calculating the square of the Euclidean distance between the original observation value of the sample and its reconstructed value under the principal component model, SPE Statistics monitor the residual space. k Pattern Statistics and The statistics are:

[0165]

[0166]

[0167] in, T represents the transpose of the matrix, Indicates the k The transpose of the pattern data, represents the projection matrix, represents the transpose of the projection matrix, Indicates the k The data of the pattern, I represents the identity matrix, Indicates the k The covariance matrix of the low-dimensional representation of the pattern training data, and , and the corresponding control limits are obtained through the kernel density estimation algorithm.

[0168] For the new sample obtained, first determine the mode to which the new sample belongs, and use the final projection matrix to calculate the new sample's Statistics and SPE Statistics;

[0169] pass Statistics and SPE Statistics can observe the difference between data samples and normal operating conditions, and compare the difference with the control limit to reflect whether the difference is within the normal allowable range.

[0170] The new sample Statistics and SPE The statistic is compared with the control limits corresponding to the mode to which the new sample belongs, and an industrial process failure message is issued when the statistic exceeds the control limits.

[0171] If the statistic exceeds the control limit of the corresponding mode, it indicates that the current data sample is significantly different from the normal operating state and exceeds the normal range, and it can be considered to be in a fault state; if the statistic does not exceed the control limit, it indicates that the current data sample is within the range allowed by the normal operating state and can be considered to be in a normal state.

[0172] It should be noted that during the training process, the above-mentioned computational model uses data samples collected under normal operating conditions for training, and manually defines the confidence level based on past experience with normal operation to determine the control limits of normal operating conditions. When the multi-mode industrial process is subsequently monitored online, information on whether a fault has occurred can be directly obtained based on the comparison results of the statistics of the newly obtained data samples with the control limits.

[0173] Example 4

[0174] Figure 4This is a schematic diagram of the structure of a multi-mode industrial process monitoring device according to a fourth embodiment of the present invention. In this embodiment, the multi-mode industrial process monitoring device includes:

[0175] Data collection module 810, for collecting and pre-processing the operating data of the multi-mode industrial process;

[0176] The data classification module 820 is used to perform pattern classification on the pre-processed operation data through cluster analysis;

[0177] The matrix calculation module 830 is used to construct the inter-mode affinity matrix and the intra-mode affinity matrix according to the operation data of each mode and calculate the divergence matrix;

[0178] A final objective function calculation module 840 is used to form a final objective function according to the inter-mode scatter matrix and the intra-mode scatter matrix;

[0179] Function iteration module 850, used for iteratively calculating the final projection matrix of the final target function;

[0180] The fault diagnosis module 860 is used to calculate the statistics and control limits of each mode according to the projection matrix, and calculate the statistics of the new sample and compare them with the control limits of the corresponding mode to issue industrial process fault information.

[0181] This embodiment preprocesses the collected multi-mode industrial process operating data through a data collection module. The data classification module clusters the preprocessed operating data to achieve pattern classification. The matrix calculation module constructs an inter-mode affinity matrix and an intra-mode affinity matrix and calculates a divergence matrix. The final objective function calculation module constructs a final objective function. The function iteration module iterates the final objective function to calculate a final projection matrix. The fault diagnosis module calculates the statistics and control limits of each mode based on the projection matrix, and compares the statistics of new samples with the control limits of their corresponding modes to issue fault information. By analyzing the local structure of the data within each mode contained in the features, the characteristics of each mode, the relationships between different modes, and the common information between multiple modes, the industrial process is analyzed to identify whether the data sample of the industrial process is a fault sample. This fully considers the characteristics of each mode and the relationships between modes, resulting in more accurate identification results and improved fault monitoring accuracy.

[0182] The multi-mode industrial process monitoring device provided by the embodiment of the present invention can execute the multi-mode industrial process monitoring method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0183] Example 5

[0184] Figure 5This is a structural diagram of a computer device according to Embodiment 5 of the present invention. Figure 5 A block diagram of an exemplary device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 5 The device 12 shown is only an example and should not bring any limitation to the functionality and scope of use of the embodiments of the present invention.

[0185] like Figure 5 As shown, device 12 is implemented as a general-purpose computing device. Components of device 12 may include, but are not limited to, one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including system memory 28 and processing unit 16).

[0186] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. Examples of these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0187] Device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by device 12, including volatile and non-volatile media, removable and non-removable media.

[0188] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be configured to read and write non-removable, non-volatile magnetic media ( Figure 5 Not shown, usually called a "hard drive"). Although Figure 5 Although not shown, a magnetic disk drive for reading and writing to a removable non-volatile magnetic disk (e.g., a "floppy disk"), as well as an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of various embodiments of the present invention.

[0189] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may include an implementation of a network environment. Program modules 42 generally implement the functions and / or methodologies of the embodiments described herein.

[0190] Device 12 may also communicate with one or more external devices 14 (e.g., a keyboard, pointing device, display 24, etc.), one or more devices that enable a user to interact with device 12 / server / computer, and / or any device that enables device 12 to communicate with one or more other computing devices (e.g., a network card, a modem, etc.). This communication may occur via input / output (I / O) interface 22. Furthermore, device 12 may communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via network adapter 20. As shown, network adapter 20 communicates with other modules of device 12 via bus 18. It should be understood that, although not shown, other hardware and / or software modules may be used in conjunction with device 12, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0191] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28 , such as implementing the multi-mode industrial process monitoring method provided by an embodiment of the present invention.

[0192] Example 6

[0193] The sixth embodiment of the present invention further provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to execute the multi-mode industrial process monitoring method provided in the above embodiment.

[0194] The computer storage medium of the embodiments of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0195] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0196] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0197] Computer program code for performing the operations of the present invention may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0198] Note that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A multi-mode industrial process monitoring method, characterized in that: include: S101, collecting operation data of a multi-mode industrial process and preprocessing the operation data to obtain preprocessed operation data; S102, performing cluster analysis on the pre-processed operation data, dividing the modes according to the clustering results, and obtaining operation data of each mode; S103, constructing an inter-mode affinity matrix and an intra-mode affinity matrix using the operation data of each mode, and calculating an inter-mode scatter matrix and an intra-mode scatter matrix based on the inter-mode affinity matrix and the intra-mode affinity matrix; S104, constructing a final objective function based on the inter-mode scatter matrix and the intra-mode scatter matrix, wherein the final objective function reflects the intra-mode information, inter-mode information, and common information of each mode industrial process, and extracting the projection matrix, the shared matrix, and the error matrix using the final objective function; S105, iteratively updating the projection matrix, the shared matrix, and the error matrix according to the final objective function to obtain the final projection matrix; S106, using the final projection matrix to calculate the statistics and control limits of each mode. When a new sample arrives, the statistics of the new sample are calculated and compared with the control limits of the corresponding mode. When the statistics exceed the control limits, an industrial process fault message is issued; The method further includes constructing a final objective function using the inter-mode scatter matrix and the intra-mode scatter matrix: in, represents the projection matrix, which is the mapping of high-dimensional data to low-dimensional data and preserves the data characteristics, and , is the real number space, m Indicates the number of variables, d represents the data dimension after dimensionality reduction, Indicates transposing the projection matrix. represents the equilibrium parameter, represents the trace of the matrix, represents the intra-mode scatter matrix, represents the inter-mode divergence matrix, It means to find the minimum value when satisfying the objective function. , represents the shared matrix, and , n represents the number of samples of a single pattern, Represents the shared matrix C of F The square of the norm, and is the penalty parameter, Indicates the i The error matrix of each mode is , Indicates the i Model error matrix The 2,1 norm of is used to make the error matrix sparse, Indicates the i Sample data of a pattern.

2. The method according to claim 1, characterized in that The S104 includes: The final objective function is constructed using the inter-mode scatter matrix and the intra-mode scatter matrix; The augmented Lagrangian function is introduced to solve the final objective function and the augmented Lagrangian function of the final objective function is obtained; The projection matrix, shared matrix and error matrix are extracted using the augmented Lagrangian function of the final objective function.

3. The method according to claim 2, characterized in that The method of constructing an objective function by using the inter-mode scatter matrix and the intra-mode scatter matrix includes: The objective function is constructed by using the inter-mode divergence matrix and the intra-mode divergence matrix and introducing the idea of ​​representation learning.

4. The method according to claim 3, characterized in that The S105 includes: According to the final objective function, the projection matrix and error matrix in the final objective function are fixed, and the shared matrix is ​​updated; Fix the projection matrix and the updated shared matrix, and update the error matrix; Fix the updated shared matrix and the updated error matrix, and update the projection matrix; The updated projection matrix is ​​compared with a preset error threshold to determine the final projection matrix.

5. The method according to claim 4, characterized in that The method further comprises: Update the Lagrange multipliers and penalty parameters based on the updated projection matrix, the updated shared matrix, and the updated error matrix.

6. The method according to claim 1, characterized in that The S103 includes: All sample data in the operation data of each mode are used to form a multi-mode industrial process matrix. The Euclidean distance between each sample in the multi-mode industrial process matrix is ​​used to construct the inter-mode affinity matrix and the intra-mode affinity matrix respectively. The inter-mode scatter matrix and the intra-mode scatter matrix are calculated based on the inter-mode affinity matrix and the intra-mode affinity matrix.

7. The method according to claim 1, characterized in that The S106 includes: Use the final projection matrix to calculate the Statistics, SPE Statistics and control limits; For the new sample obtained, first determine the mode to which the new sample belongs, and use the final projection matrix to calculate the new sample's Statistics and SPE Statistics; The new sample Statistics and SPE The statistic is compared with the control limits corresponding to the mode to which the new sample belongs, and an industrial process failure message is issued when the statistic exceeds the control limits.

8. A multi-mode industrial process monitoring device, characterized in that: include: Data collection module, used to collect and pre-process the operating data of multi-mode industrial processes; Data classification module, used to perform pattern classification on pre-processed operation data through cluster analysis; A matrix calculation module is used to construct an inter-mode affinity matrix and an intra-mode affinity matrix according to the operation data of each mode and calculate a divergence matrix; A final objective function calculation module, used for forming a final objective function according to the inter-mode scatter matrix and the intra-mode scatter matrix; Function iteration module, used to iteratively calculate the final projection matrix of the final objective function; The fault diagnosis module is used to calculate the statistics and control limits of each mode according to the projection matrix, and to calculate the statistics of the new sample and compare them with the control limits of the corresponding mode to issue industrial process fault information; Among them, the final objective function is: in, represents the projection matrix, which is the mapping of high-dimensional data to low-dimensional data and preserves the data characteristics, and , is the real number space, m Indicates the number of variables, d represents the data dimension after dimensionality reduction, Indicates transposing the projection matrix. represents the equilibrium parameter, represents the trace of the matrix, represents the intra-mode scatter matrix, represents the inter-mode divergence matrix, It means to find the minimum value when satisfying the objective function. , represents the shared matrix, and , n represents the number of samples of a single pattern, Represents the shared matrix C of F The square of the norm, and is the penalty parameter, Indicates the i The error matrix of each mode is , Indicates the i Model error matrix The 2,1 norm of is used to make the error matrix sparse, Indicates the i Sample data of a pattern.

9. A computer device, characterized in that: The device comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-mode industrial process monitoring method according to any one of claims 1 to 7.

10. A storage medium comprising computer executable instructions, wherein the computer executable instructions, when executed by a computer processor, are used to perform the multi-mode industrial process monitoring method according to any one of claims 1 to 7.

Images