A Distributed Process Monitoring Method and Device for Intra-block and Inter-block Collaborative Modeling

Abstract

The present invention relates to the technical field of industrial process monitoring. More specifically, it relates to a distributed process monitoring method and device for collaborative modeling within and between blocks. This method includes: The offline modeling stage includes the following steps: decomposing the process data to obtain each sub-block; performing slow feature analysis method modeling within each sub-block; performing canonical correlation analysis modeling between sub-blocks; The online monitoring stage includes the following steps: partitioning the new sampled data; substituting the partitioned sampled data into the slow feature analysis method model within the sub-block and the canonical correlation analysis model between sub-blocks respectively to obtain the corresponding feature components; using Bayesian inference fusion to obtain a comprehensive index as the final monitoring statistic and comparing it with the corresponding control limit to detect the state of the current sub-block. The present invention can detect faults from different characteristic perspectives of the data, provide a preliminary range for fault location, so as to improve the detection performance of the process and the fault detection rate.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process monitoring. More specifically, it relates to a distributed process monitoring method and device for collaborative modeling within and between blocks. Background Art

[0002] In modern industry, ensuring the safe operation of process production, the quality of products, and improving production efficiency have always been the directions of research. From the most traditional model-based methods, which rely on establishing accurate mathematical models of the process (such as physical laws and chemical reaction processes) to monitor whether there are deviations in the actual system production status. However, with the increasing complexity of process systems, it has become more difficult and costly to establish characteristic models.

[0003] People accumulate relevant knowledge and expert experience in industrial processes to form knowledge-based methods for process monitoring. The monitoring results of knowledge-based methods are more intuitive and understandable. However, so far, in some process fields, the time for accumulating expert experience and knowledge is short and the knowledge base is small. Therefore, knowledge-based methods are difficult to apply.

[0004] Different from the need for long-term accumulation of knowledge, with the development of modern industrial systems, a large amount of data containing industrial process information can be obtained. Therefore, the research on data-driven process monitoring methods has been carried out vigorously.

[0005] The core of data-driven process monitoring is the mining of data characteristics. In fact, for different industrial processes, their process data exhibit different characteristics or have multiple characteristics. The main characteristics of common process industry data are as follows:

[0006] High dimensionality. The high dimensionality of data is inevitable. The development trend of modern industrial processes is towards large scale and

[0007] complexity. The entire production process consists of various components and operating devices. In each part, there are some important variables that can be measured. Therefore, the dimensionality of the data collected during the process is getting higher and higher.

[0008] Non-Gaussian distribution. Most of the classic process monitoring methods are used on the premise of Gaussian processes. However, in actual production, not all variables follow a Gaussian distribution. Due to non-Gaussian noise or feedback control loops, etc., some variables may follow different types of non-Gaussian distributions. Therefore, many monitoring methods for corresponding non-Gaussian process data have also been developed.

[0009] Nonlinearity. The existence of nonlinearity is very common. In a process, some process variables are linearly correlated, while others are not. There may also be a non - linear correlation between process variables and quality variables. There are diverse non - linear relationships in process data. Analyze according to specific characteristics and establish corresponding process monitoring models.

[0010] Time - variability. In the industrial production process, its operating state is affected by many aspects and changes over time, such as fluctuations in feed components, aging of process equipment, scheduling of production plans, etc. For this reason, corresponding solutions, such as process monitoring schemes of adaptive technology, multi - modal methods, etc., have also been proposed.

[0011] Autocorrelation. In addition to the correlation between variables, there is also dynamics within the variables themselves. The current measured value of a variable is autocorrelated with its past values. The time - series correlation of variables is related not only to the nature of the process but also may be related to control loops, feedback loops, process noise, etc. Some faults will affect the dynamics of the process, so the exploration of data dynamics is also an important part of process monitoring.

[0012] Nowadays, the complexity of industrial systems is getting higher and higher, the collected data is getting more and more, and the characteristics of the data are getting more and more complex. How to better explore data characteristics is the key to improving the performance of process monitoring.

[0013] Distributed canonical correlation analysis (CCA) process monitoring based on partial sub - block communication. For large - scale processes, the process is decomposed in combination with the actual process background, and a topological matrix used to describe the connection between sub - blocks is proposed. According to the topological matrix, CCA modeling is carried out for connected and information - interacting sub - blocks, the correlation between the process data of two sub - blocks is explored, and process monitoring is realized by observing the change of the space constructed based on the residual vector.

[0014] However, the distributed CCA process monitoring method based on partial sub - block communication only focuses on exploring the correlation between sub - blocks. The measurement data of each sub - block also contains a lot of process information in the current range, and features can be extracted from the local data to construct a principal component feature space for supervision.

[0015] Moreover, in the distributed CCA modeling based on partial sub - block communication, a serious problem may be faced. When a certain sub - block only communicates with a certain sub - block in the topological structure, if there are problems in data transmission, such as line interruption, etc., the data of the current sub - block cannot be used for process monitoring.

[0016] Therefore, a method to solve the above problems needs to be proposed. Summary of the Invention

[0017] The object of the present invention is to provide a distributed process monitoring method and device for collaborative modeling within and between blocks, so as to solve the problem of poor performance of the distributed CCA process monitoring in the prior art.

[0018] To achieve the above object, the present invention provides a distributed process monitoring method for collaborative modeling within and between blocks, which includes two stages: offline modeling and online monitoring. The offline modeling stage specifically includes the following steps:

[0019] Step S1: Decompose the process data to obtain each sub-block, and construct a topological matrix C;

[0020] Step S2: Model each sub-block using the slow feature analysis method within the sub-block, and extract the slow features and retain the projection matrix;

[0021] Step S3: Calculate the statistics respectively according to the extracted slow features, and obtain the corresponding control limits through kernel density estimation;

[0022] Step S4: Perform canonical correlation analysis modeling between connected sub-blocks according to the topological matrix C, and extract the canonical correlation components and retain the projection matrix;

[0023] Step S5: Generate a residual vector according to the canonical correlation components and calculate the statistics, and obtain the corresponding control limits through kernel density estimation;

[0024] The online monitoring stage specifically includes the following steps:

[0025] Step S6: For the new sampled data, perform corresponding partitioning according to Step S1, and decompose it to obtain each sub-block;

[0026] Step S7: Substitute the partitioned sampled data into the slow feature analysis method model within the sub-block and the canonical correlation analysis model between the sub-blocks respectively to obtain the corresponding feature components;

[0027] Step S8: Calculate the statistics of the models corresponding to the slow feature analysis method model and the canonical correlation analysis model respectively;

[0028] Step S9: For the different detection results of multiple models of the same sub-block, use Bayesian inference fusion to obtain a comprehensive index ET 2 as the final monitoring statistic, and compare it with the corresponding control limit to detect the state of the current sub-block;

[0029] Step S10: Calculate the fault detection index FD of the entire process, and compare it with the corresponding control limit. When it exceeds the control limit, it indicates that there is a fault in the process, otherwise it is considered that the operating state of the process is normal.

[0030] In one embodiment, the above Step S1 further includes:

[0031] Decompose the process data by combining the process background and mechanism knowledge. For each obtained sub-block, the corresponding expression is:

[0032] X = [X1, X2, …, X i …,, X B

[0033] where X i is the i-th sub-block, and B is the number of sub-blocks.

[0034] In one embodiment, in step S2, for the slow features extracted from each sub-block, the corresponding expression is:

[0035] s i = W i X i

[0036] where s i is the slow feature of the i-th sub-block, W i is the weight matrix corresponding to the i-th sub-block, and X i is the i-th sub-block.

[0037] In one embodiment, for the statistic in step S3, the corresponding expression is:

[0038]

[0039] where is the statistic corresponding to the i-th sub-block, used to monitor the change of the components representing the essence of the process state in the subspace, and s i is the slow feature of the I-th sub-block.

[0040] In one embodiment, the kernel density estimation corresponding expression is:

[0041]

[0042] where y is the data to be estimated, y i is the observed value of the process data, n is the number of samples, h is the smoothing parameter, and K(·) is the kernel function.

[0043] In one embodiment, in step S4, perform canonical correlation analysis modeling between the sub-block and the sub-blocks with which it has information interaction according to the topological matrix C. The expression of the obtained canonical correlation components is:

[0044]

[0045]

[0046] where A ij ​and B ij are the projection matrices of sub - block i and sub - block j, and u ij and v ij are the canonical components of sub - block i and sub - block j respectively.

[0047] In one embodiment, in the step S5, the residual vector established based on the extracted canonical correlation components has the corresponding expression:

[0048]

[0049] where r ij is the residual vector between sub - block i and sub - block j, Λ k,ij is the diagonal matrix of the canonical correlation coefficients of the first k pairs of correlated variables between sub - block i and sub - block j, X i is the i - th sub - block, and X j is the j - th sub - block;

[0050] In the step S5, the statistic constructed based on the residual vector has the corresponding expression:

[0051]

[0052] where is the statistic constructed based on the residual vector between sub - block i and sub - block j, and I k is the K - order identity matrix.

[0053] In one embodiment, the final monitoring statistic ET 2 in the step S9 has the corresponding expression:

[0054]

[0055] where is the final monitoring statistic of the i - th sub - block, x new is the online data, is the conditional probability of the abnormal operating state of the process, is the probability of failure occurrence, and P(c ij ) = c ij is the coefficient representing the connection relationship between sub - blocks based on the topological matrix C.

[0056] In one embodiment, in the step S9, the control limit CL i of the final monitoring statistic of the i - th sub - block is the prior probability (1 - α) of the occurrence of an abnormal working condition of the process.

[0057] In one embodiment, in the step S10, the expression of the fault detection index FD of the whole process is:

[0058]

[0059] Among them, the control limit of the fault detection index FD is 1.

[0060] To achieve the above object, the present invention provides a distributed process monitoring device for intra-block and inter-block collaborative modeling, including:

[0061] A memory for storing instructions executable by a processor;

[0062] A processor for executing the instructions to implement the method as described in any one of the above.

[0063] To achieve the above object, the present invention provides a computer-readable medium having computer instructions stored thereon, wherein when the computer instructions are executed by a processor, the method as described in any one of the above is executed.

[0064] The distributed process monitoring method and device for intra-block and inter-block collaborative modeling provided by the present invention can provide process monitoring, detect faults from different characteristics of data, and provide a preliminary range for fault location, so as to improve the detection performance of the process and the fault detection rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The above and other features, properties, and advantages of the present invention will become more apparent from the following description in conjunction with the drawings and embodiments, in which the same reference numerals in the drawings always represent the same features, wherein:

[0066] Figure 1 Discloses a flowchart of a distributed process monitoring method for intra-block and inter-block collaborative modeling according to an embodiment of the present invention;

[0067] Figure 2 Discloses a schematic diagram of the principle of a distributed process monitoring method for intra-block and inter-block collaborative modeling according to an embodiment of the present invention;

[0068] Figure 3 Discloses a flowchart of the No. 1 reference simulation model according to an embodiment of the present invention;

[0069] Figure 4 Discloses a schematic diagram of the fault detection result of fault 3 in the BMS1 process based on the SFA method;

[0070] Figure 5 Discloses a schematic diagram of the fault detection result of fault 3 in the BMS1 process based on the distributed CCA process monitoring method using partial sub-block communication;

[0071] Figure 6 Discloses a schematic diagram of the fault detection result of fault 3 in the BMS1 process according to an embodiment of the present invention;

[0072] Figure 7Reveals a schematic diagram of the fault detection result of fault 3 in the TE process by sub-block 1 in the distributed CCA process monitoring method based on partial sub-block communication;

[0073] Figure 8 Reveals a schematic diagram of the fault detection result of fault 16 in the TE process by sub-block 1 according to an embodiment of the present invention;

[0074] Figure 9 Reveals a schematic diagram of the fault detection result of fault 3 in the TE process by sub-block 2 in the distributed CCA process monitoring method based on partial sub-block communication;

[0075] Figure 10 Reveals a schematic diagram of the fault detection result of fault 16 in the TE process by sub-block 2 according to an embodiment of the present invention;

[0076] Figure 11 Reveals a schematic diagram of the fault detection result of fault 3 in the TE process by sub-block 3 in the distributed CCA process monitoring method based on partial sub-block communication;

[0077] Figure 12 Reveals a schematic diagram of the fault detection result of fault 16 in the TE process by sub-block 3 according to an embodiment of the present invention. Detailed implementation manners

[0078] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the invention and are not used to limit the invention.

[0079] To enable those skilled in the art to understand the features and effects of the present invention, the following provides a general description and definition of the terms and expressions mentioned in the specification and claims. Unless otherwise specified, all technical and scientific terms used herein shall have the ordinary meaning understood by those skilled in the art for the present invention. In case of conflict, the definition in this specification shall prevail.

[0080] In this article, for the sake of brevity of description, all possible combinations of all technical features in each embodiment or example are not described. Therefore, as long as there is no contradiction in the combination of these technical features, the technical features in each embodiment or example can be combined arbitrarily, and all possible combinations should be considered to be within the scope described in this specification.

[0081] Figure 1 Reveals a flow chart of the distributed process monitoring method for intra-block and inter-block collaborative modeling according to an embodiment of the present invention, Figure 2 Reveals a schematic diagram of the principle of the distributed process monitoring method for intra-block and inter-block collaborative modeling according to an embodiment of the present invention, asFigure 1 and Figure 2 As shown in and

[0082] , the present invention proposes a distributed process monitoring method for intra-block and inter-block collaborative modeling, including two stages: offline modeling and online monitoring. The offline modeling stage specifically includes the following steps:

[0082] Step S1: Decompose the process data to obtain each sub-block, and construct a topological matrix C;

[0083] Step S2: Perform slow feature analysis (SFA) modeling within each sub-block, and extract slow features and retain the projection matrix;

[0084] Step S3: Calculate statistics based on the extracted slow features respectively, and obtain the corresponding control limits through kernel density estimation;

[0085] Step S4: Perform canonical correlation analysis (CCA) modeling between connected sub-blocks according to the topological matrix C, and extract canonical correlation components and retain the projection matrix;

[0086] Step S5: Generate a residual vector based on the canonical correlation components and calculate statistics, and obtain the corresponding control limits through kernel density estimation;

[0087] The online monitoring stage specifically includes the following steps:

[0088] Step S6: For the new sampled data, perform corresponding partitioning according to Step S1 to decompose it into each sub-block;

[0089] Step S7: Substitute the partitioned sampled data into the slow feature analysis (SFA) model within the sub-block and the canonical correlation analysis (CCA) model between sub-blocks respectively to obtain the corresponding feature components;

[0090] Step S8: Calculate the statistics of the models corresponding to the slow feature analysis method model and the canonical correlation analysis model respectively;

[0091] Step S9: For the different detection results of multiple models of the same sub-block, use Bayesian inference fusion to obtain a comprehensive index ET 2 as the final monitoring statistic, and compare it with the corresponding control limit to detect the state of the current sub-block;

[0092] Step S10: Calculate the fault detection index FD of the entire process, and compare it with the corresponding control limit. When it exceeds the control limit, it indicates that there is a fault in the process, otherwise it is considered that the operating state of the process is normal.

[0093] The present invention proposes a distributed process monitoring method for intra-block and inter-block collaborative modeling, namely, a distributed process monitoring method for intra-block and inter-block collaborative modeling based on SFA and CCA (PDCCA-SFA). Based on the distributed CCA process monitoring using partial sub-block communication, this method combines the SFA method to establish an SFA local model within each sub-block. That is, for each sub-block, a local model is established using the locally measured and stored data for supervision, while a CCA model is established for the information-interacting sub-blocks to detect faults from different characteristics of the data, improving the detection performance of the process. The SFA method (slow feature analysis method) is different from the existing principal component analysis technology (PCA) method. SFA extracts the components that change most slowly in the data representing the essence of the process and also considers the process dynamics. Since the SFA algorithm explores the dynamic changes of the process from the internal perspective of the data, the extracted slow features contain not only the steady-state information of the process but also the dynamic information of the process, and can be used for dynamic process modeling for anomaly detection.

[0094] The principle of the SFA method is specifically as follows:

[0095] Suppose there is a p-dimensional input signal x(t) = (x1(t), x2(t), …, x p (t)) T , the SFA method hopes to find a transformation function f(x) = (f1(x), f2(x), …, f p (x)) T , which can transform the input signal into a series of output signals s(t) = f(x(t)) that change as slowly as possible. Its optimization objective is:

[0096]

[0097] where, represents the first derivative of the output signal s j (t), calculated by , <·> t represents the time average, Δs j (t) represents the change rate of the output signal s j (t), <s j (t)> t = 0 and respectively represent the constraints of zero mean and unit variance to avoid obtaining trivial solutions to ensure that the output signals are uncorrelated with each other and are output in a certain order, that is, the change of the first output signal is the slowest, and the second output signal is the second slowest changing.

[0098] The SFA method is a linear method, which assumes that there is a correlation between the input variables and the output variables. Therefore, the extracted slow feature s j(t) can be regarded as a linear combination of a series of input signals Therefore, the expression of the extracted slow feature is:

[0099] s(t) = Wx(t)

[0100] where W = [w1, w2, …, w p T is the weight matrix

[0101] Given the input vector x, before applying the slow feature analysis algorithm, it is necessary to perform whitening processing to eliminate the correlation between variables. The whitening processing is carried out through singular value decomposition (SVD). B = <xx T > t represents the expectation of the covariance matrix of the input vector x. Perform singular value decomposition on it:

[0102] B = UΛU T

[0103] where Λ is the diagonal matrix composed of eigenvalues, and U is the matrix composed of the corresponding eigenvectors. Thus, the whitened data z can be obtained:

[0104] z = Λ -1 / 2 U T x = Qx

[0105] satisfying cov(z) = <zz T > = I p , I p is the identity matrix, and Q = Λ -1 / 2 U T is the whitening matrix

[0106] Therefore, the optimization objective of the SFA algorithm can be simplified to finding a matrix P that satisfies the following transformation:

[0107] s = Pz

[0108] while satisfying

[0109] ss T = I p

[0110] It is deduced that <PP T > = I p , and the matrix P is an orthogonal matrix

[0111] Correspondingly, in order to extract slow features, perform singular value decomposition on the covariance matrix of the first derivative of the whitened data z:

[0112] ​

[0113] Thus, the expressions for the weight matrix W and the slow feature s can be obtained as follows:

[0114] W = PQ

[0115] s = Wx.

[0116] Next, in combination with Figure 1 and Figure 2 , the above steps of the present invention will be described in detail. It should be understood that within the scope of the present invention, the above technical features of the present invention and the technical features specifically described below (such as in the embodiments) can be combined with each other and are interrelated, so as to constitute a preferred technical solution.

[0117] Step S1: Decompose the process data to obtain each sub-block, and construct the topological matrix C.

[0118] Combining the process background and mechanism knowledge to decompose the process, the obtained each sub-block has the corresponding expression as follows:

[0119] X = [X1, X2,..., X i …,, X B B

[0120] wherein, X i is the i-th sub-block, and B is the number of sub-blocks.

[0121] Based on the method of dividing blocks according to process knowledge, the difficulty of division for different production processes is different. Such division has a certain feasibility, and it is also beneficial for preliminary fault range positioning after a fault is detected.

[0122] After dividing the process into B sub-blocks according to mechanism knowledge and expert experience, define the corresponding process topological matrix C. The topological matrix C is proposed to describe the connection relationship between sub-blocks. Considering that in actual industrial processes, the structures of different process flows are also diverse, not all sub-blocks can exchange information with each other, and not all sub-blocks have internal connections such as hierarchical progression in structure.

[0123] Therefore, compared with the distributed CCA process monitoring where all sub-blocks interact pairwise, the present invention proposes a topological matrix C to describe the connection mode between sub-blocks, and based on the topological matrix C, model some of the sub-blocks that communicate, thereby further reducing the communication load of data transmission, and at the same time, the established distributed CCA process monitoring is more in line with the actual industrial process.

[0124] Each element in the topological matrix C represents whether there is information interaction between the corresponding two sub-blocks. If so, set it to 1, and if not, set it to 0. Since the element represents the relationship between two sub-blocks, the elements on the diagonal are all set to 0.

[0125] Step S2: Perform slow feature analysis method modeling within each sub-block, and extract slow features and retain the projection matrix.

[0126] The expression for the slow features extracted from each sub-block is:

[0127] s i = W i X i

[0128] where s i is the slow feature of the i-th sub-block, and W i is the weight matrix corresponding to the i-th sub-block.

[0129] Step S3: Calculate the statistics based on the extracted slow features respectively, and obtain the corresponding control limits through kernel density estimation.

[0130] The expression for the statistics calculated based on the extracted slow features respectively is:

[0131]

[0132] where is the statistic corresponding to the i-th sub-block, used to monitor the change of the components representing the essence of the process state in the subspace, and its corresponding control limit is obtained by kernel density estimation.

[0133] The statistic constructed based on the slow features can monitor the change of the components representing the essence of the process state in the subspace, thereby realizing fault detection, and its corresponding control limit can be obtained by kernel density estimation.

[0134] Kernel Density Estimation (KDE), as a non-parametric method, is often used to infer the distribution characteristics of data. The mathematical description of kernel density estimation is:

[0135]

[0136] where is the kernel density estimation, y is the data to be estimated, y i is the observed value of the process data, n is the number of samples, h is the smoothing parameter, and K(·) is the kernel function.

[0137] There are many choices for the kernel function, and the commonly used kernel function is the Gaussian function:

[0138]

[0139] Step S4: According to the topological matrix C, perform canonical correlation analysis modeling between the connected sub-blocks, and extract the canonical correlation components and retain the projection matrix.

[0140] Perform CCA modeling on the sub-block and the sub-block with which it has information interaction according to the topological matrix C. The expression of the obtained canonical component is:

[0141]

[0142]

[0143] where A ij and B ij are the projection matrices of sub-block i and sub-block j, and u ij and v ij are the canonical components of sub-block i and sub-block j respectively.

[0144] Step S5: Generate a residual vector according to the canonical correlation components and calculate the statistic, and obtain the corresponding control limit through kernel density estimation.

[0145] The expression of the residual vector established based on the extracted canonical components is:

[0146]

[0147] where r ij is the residual vector between sub-block i and sub-block j, and Λ k,ij is the diagonal matrix of the canonical correlation coefficients of the first k pairs of correlated variables between sub-block i and sub-block j;

[0148] The expression of the statistic constructed based on the residual vector is:

[0149]

[0150] where is the statistic constructed based on the residual vector between sub-block i and sub-block j, and I k is the K-order identity matrix.

[0151] The statistic constructed based on the residual vector is used to monitor the change of the features reflecting the relationship between sub-blocks in the subspace, so as to realize fault detection, and its corresponding control limit can be obtained by kernel density estimation.

[0152] The calculation methods corresponding to steps S6 to S8 in the online monitoring stage are the same as those in the offline modeling stage, and will not be elaborated here.

[0153] Step S9: For the different detection results of multiple models of the same sub-block, use Bayesian inference fusion to obtain the comprehensive index ET 2As the final monitoring statistic, it is compared with the corresponding control limit to detect the status of the current sub-block.

[0154] For the same sampling point, the SFA model and the corresponding CCA model of the current sub-block may have different detection results. Bayesian inference is used to fuse the detection results of the current sub-block to obtain the final monitoring statistic ET of each sub-block. 2 。

[0155] The final monitoring statistic ET 2 , and the corresponding expression is:

[0156]

[0157] where is the final monitoring statistic (i.e., the comprehensive index) of the i-th sub-block, x new is the online data (i.e., the new query sample), is the conditional probability of the abnormal operating state of the process, is the probability of failure occurrence, P(c ij ) = c ij is the coefficient representing the connection relationship between sub-blocks based on the topological matrix C;

[0158] The conditional probability has the following expression:

[0159]

[0160] where is the statistic of the i-th sub-block corresponding control limit;

[0161] The probability of failure occurrence has the following expression:

[0162]

[0163] where is the prior probability of the abnormal operating state, is the probability corresponding to the online data, is the prior probability of the normal operating state.

[0164] The control limit CL corresponding to the final monitoring statistic of the i-th sub-block i is the prior probability (1 - α) of the occurrence of abnormal working conditions in the process. When the final monitoring statistic of the i-th sub-block is higher than its corresponding control limit, it is judged that the sub-block is abnormal. Otherwise, it is considered that the sub-block is operating under normal working conditions.

[0165] Step S10: Calculate the fault detection index FD for the entire process and compare it with the corresponding control limit. When the control limit is exceeded, it indicates that a fault has occurred in the process; otherwise, the process is considered to be in a normal operating state.

[0166] Since, from the perspective of monitoring the entire process, a fault in any sub-block will affect the entire production process and an alarm should be given from a global perspective, an overall index is designed to represent the monitoring situation of the entire process. The expression for the fault detection index FD of the entire process is as follows:

[0167]

[0168] Among them, according to the definition of the fault detection index FD, its control limit is 1.

[0169] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and the detailed implementation manner and specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0170] In this embodiment, the distributed process monitoring method based on intra-block and inter-block collaborative modeling of SFA and CCA proposed by the present invention is applied to the urban sewage treatment process and compared and analyzed with the SFA and the distributed CCA method based on partial sub-block communication.

[0171] The urban sewage treatment process is the general Benchmark Simulation Model No. 1 (BSM1). The BSM1 benchmark simulation model is a sewage treatment process model proposed by the EU Science and Technology Cooperation Organization and is widely used in the research of sewage treatment processes by personnel and institutions around the world.

[0172] In fact, the BSM1 model mainly consists of two models, namely the activated sludge process model and the double exponential sedimentation model. The BSM1 benchmark simulation model mainly focuses on removing carbon and nitrogen in sewage, so that the treated wastewater can meet the secondary discharge standard.

[0173] Figure 3 Reveals the flowchart of the Benchmark Simulation Model No. 1 according to an embodiment of the present invention, as Figure 3 shown, the process structure of BSM1 consists of a bioreactor and a secondary sedimentation tank. The bioreactor consists of 5 well-mixed units. Among them, the first 2 units are anaerobic closed containers (referred to as anoxic tanks or anaerobic tanks), and the last 3 units are aerobic exposed containers (referred to as aeration tanks or aerobic tanks).

[0174] In this model, organic matter is mainly consumed through the vital activities of microorganisms, while nitrogen is removed through nitrification by microorganisms in the aerobic zone and denitrification in the anaerobic zone.

[0175] The total height of this secondary sedimentation tank is 4 meters, divided into 10 equal-height layers, and no chemical reactions occur in the secondary sedimentation tank. It is mainly a physical filtration process.

[0176] The water in the reaction unit flows out from the fifth reaction tank. Part of it flows into the sixth layer of the secondary sedimentation tank, and the other part is recycled back to the inlet of the reaction unit through a pipeline to complete the internal cycle. In the secondary sedimentation tank, the wastewater that meets the sewage discharge standards after sedimentation is discharged from the top of the secondary sedimentation tank into the river, while part of the sedimented sludge at the bottom is sent back to the first reaction tank for sludge recycling, and the other part is discharged for landfill.

[0177] The BSM1 model introduces three faults, as shown in Table 1. The first fault is that the maximum growth rate of autotrophic bacteria in the first aerobic reaction tank undergoes a step change, decreasing from 0.5 to 0.3. Due to the weakened vital activities of autotrophic bacteria, it will affect the complex biochemical reactions in the urban sewage treatment process, thus changing the state of the entire system.

[0178] The second fault is that the oxygen transfer coefficient of biological reaction tank 4 drops to half of its original value. The oxygen transfer coefficient has a great relationship with the concentration of dissolved oxygen in the reaction tank, and the concentration of dissolved oxygen is an important element in the reaction of the urban sewage treatment process, which is related to the sewage treatment effect. Because whether it is the microbial activities for removing organic matter or the denitrification of heterotrophic bacteria, they all require the participation of oxygen.

[0179] The third fault is that the SNO_2 sensor in biological reaction tank 2 has a deviation of +0.5. The SNO_2 sensor is used to measure the concentrations of nitrate nitrogen and nitrite nitrogen in biological reaction tank 2, and the measured values are input into the controller. The controller adjusts the size of the internal reflux according to the deviation to make the system operate normally. Therefore, if the measured values are inaccurate, it will affect the water purification effect and cause the wastewater not to meet the discharge standards.

[0180] Table 1 Three faults in the BSM1 process

[0181] Fault Description Type 1 Maximum growth rate of autotrophic bacteria in the first aerobic reaction tank Step 2 Oxygen transfer coefficient of biological reaction tank 4 Step 3 SNO_2 measurement value of biological reaction tank 2 Step

[0182] First, according to the sewage treatment process and structure of the BSM1 model, the 15 process measurement data used for monitoring are divided into 3 sub-blocks. Sub-block 1 is the relevant water quality measurement variables of the treated effluent, sub-block 2 is the water quality component measurement variables of the sewage inlet and reaction tank 3, and sub-block 3 is the measurement variables of reaction tanks 4 and 5. The measurement variables and sub-block divisions are shown in Tables 2 and 3 respectively:

[0183] Table 2 Measurement Variables of the BSM1 Process

[0184]

[0185]

[0186] Table 3 Sub - block Division of the BSM1 Process Variables

[0187]

[0188] After dividing the process data into sub - blocks, according to process knowledge, the expression of its topological matrix C is as follows:

[0189]

[0190] According to the urban sewage treatment process, the influent water flows through two anaerobic ponds and three aerobic ponds respectively, removes organic matter and nitrogen, and then enters the secondary sedimentation tank. After clarification, the effluent is discharged into the river. Therefore, the topological coefficients between sub - block 2 representing the influent and the third reaction tank and sub - block 3 representing the fourth and fifth reaction tanks are set to 1, and the topological coefficient from sub - block 3 to sub - block 1 representing the effluent water quality components is also 1. Therefore, in this embodiment, 2 CCA models are established.

[0191] Set the number of canonical correlation components to 3, and the prior probability α of the normal operation state of the process to 99%.

[0192] As shown in Table 4, the detection results of three faults in the BSM1 process by process monitoring based on SFA, distributed CCA process monitoring (PDCCA) of partial sub - block communication, and distributed process monitoring based on intra - block and inter - block collaborative modeling (PDCCA - SFA) are shown. It can be seen from Table 4 that for the second fault, the oxygen transfer coefficient of the 4th biological reaction tank drops to half of the original value, and all three methods can detect the abnormality;

[0193] For the first fault, the maximum growth rate of autotrophic bacteria in the first aerobic reaction tank has a step change. The method based on SFA can detect the fault, while the detection effects of the CCA method based on partial sub - block connection and the proposed intra - block and inter - block collaborative modeling method are average, only 60% and 65% respectively. The reason for the analysis may be that the influence range of this fault is relatively wide, so the global SFA model can have a good detection effect, while the detection effect of the local model based on sub - blocks is poor;

[0194] For the third fault, the SNO_2 sensor in biological reaction tank 2 has an offset. The process monitoring based on SFA and the monitoring method based on intra - block and inter - block collaborative modeling both have good detection results, and the fault detection rates are 96% and 95% respectively, while the distributed CCA monitoring method based on partial sub - block communication can hardly detect the fault, indicating that SFA can effectively detect this fault.

[0195] Table 4 Fault detection rates of the BSM1 process for SFA, PDCCA, and PDCCA - SFA

[0196]

[0197] Taking the third fault as an example, further analysis is as follows:

[0198] The third fault is specifically a 0.5 offset of the SNO_2 sensor in bioreactor 2.

[0199] Figures 4 to 6 Respectively reveal the fault detection results of three process detection methods for fault 3 of the BMS1 process, as Figure 4 and Figure 6 shown. The monitoring method based on SFA and the monitoring method of the present invention based on intra - block and inter - block collaborative modeling (PDCCA - SFA) can both detect the fault and issue an alarm after the fault occurs. As Figure 5 shown, the distributed CCA process monitoring method based on partial sub - block communication (PDCCA) can hardly detect the fault.

[0200] Comparing the fault detection results of the sub - blocks of PDCCA and PDCCA - SFA for fault 3 of the TE process, Figure 7 、 Figure 9 and Figure 11 respectively reveal the detection results of sub - blocks 1 - 3 in PDCCA, Figure 8 、 10 and Figure 12 respectively reveal the detection results of sub - blocks 1 - 3 in the monitoring method of the present invention based on intra - block and inter - block collaborative modeling (PDCCA - SFA). It can be seen that none of the three sub - blocks of the distributed CCA monitoring method based on partial sub - block communication can detect the fault, while sub - block 1 of the distributed process monitoring method based on intra - block and inter - block collaborative modeling can detect the fault, indicating that establishing a local SFA model for each sub - block can improve the monitoring performance of the process.

[0201] Because the third fault is that the sensors for measuring the concentrations of nitrate nitrogen and nitrite nitrogen are offset, resulting in a change in the size of the internal reflux regulated by the controller, affecting the sewage treatment effect. Therefore, the local SFA model of sub - block 1 where the monitoring variable is the effluent water quality components can detect the fault.

[0202] The distributed process monitoring method and device based on intra-block and inter-block collaborative modeling proposed by the present invention. After the process is decomposed, in addition to using the local measurement data of each sub-block to establish the CCA model between sub-blocks, the rich process information contained therein can be mined through slow feature analysis and used to detect faults in the current range. Therefore, after the process is decomposed, in addition to establishing the CCA model between sub-blocks according to the topological matrix, a local SFA model is established for each sub-block to detect faults in specific units, and then the detection results of the intra-block SFA model and the inter-block CCA model of each sub-block are integrated through Bayesian inference to achieve process monitoring, and the effectiveness of the method proposed by the present invention is verified through the BSM1 process simulation experiment.

[0203] The present invention proposes a distributed process monitoring device for intra-block and inter-block collaborative modeling. The distributed process monitoring device for intra-block and inter-block collaborative modeling may include an internal communication bus, a processor, a read-only memory (ROM), a random access memory (RAM), a communication port, and a hard disk. The internal communication bus can realize data communication between components of the distributed process monitoring device for intra-block and inter-block collaborative modeling. The processor can make judgments and issue prompts. In some embodiments, the processor may be composed of one or more processors.

[0204] The communication port can realize data transmission and communication between the distributed process monitoring device for intra-block and inter-block collaborative modeling and external input / output devices. In some embodiments, the distributed process monitoring device for intra-block and inter-block collaborative modeling can send and receive information and data from the network through the communication port. In some embodiments, the distributed process monitoring device for intra-block and inter-block collaborative modeling can perform data transmission and communication with external input / output devices in a wired form through the input / output terminal.

[0205] The distributed process monitoring device for intra-block and inter-block collaborative modeling may also include program storage units and data storage units in different forms, such as a hard disk, a read-only memory (ROM), and a random access memory (RAM), which can store various data files used for computer processing and / or communication, as well as possible program instructions executed by the processor. The processor executes these instructions to implement the main part of the method. The results processed by the processor are transmitted to an external output device through the communication port and displayed on the user interface of the output device.

[0206] For example, the implementation process file of the above-mentioned distributed process monitoring device for intra-block and inter-block collaborative modeling may be a computer program, stored in the hard disk and can be recorded into the processor for execution to implement the method of the present invention.

[0207] When the implementation process file of the distributed process monitoring method for in-block and inter-block collaborative modeling is a computer program, it can also be stored in a computer-readable storage medium as an article. For example, the computer-readable storage medium may include, but is not limited to, magnetic storage devices (e.g., hard disks, floppy disks, magnetic strips), optical discs (e.g., compact discs (CDs), digital versatile discs (DVDs)), smart cards, and flash memory devices (e.g., electrically erasable programmable read-only memories (EPROMs), cards, sticks, key drives). In addition, the various storage media described herein can represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable medium" may include, but is not limited to, wireless channels and various other media (and / or storage media) that can store, contain, and / or carry code and / or instructions and / or data.

[0208] Compared with the prior art, the present invention provides a distributed process monitoring method and device for in-block and inter-block collaborative modeling, which has the following beneficial effects:

[0209] 1) Based on the distributed CCA process monitoring based on partial sub-block communication, the SFA method is combined to establish an SFA local model within each sub-block, and a CCA model is established between the sub-blocks with information interaction to detect faults from different characteristics of the data, so as to improve the detection performance of the process and the fault detection rate;

[0210] 2) By decomposing a high-dimensional industrial system into several low-dimensional sub-blocks and establishing data-driven models respectively, the computational complexity is reduced, and the division based on prior knowledge and process mechanism is more reliable, and it can better monitor the state of large-scale processes;

[0211] 3) Using in-block modeling is more conducive to detecting some small faults hidden in high-dimensional data and has stronger learning ability;

[0212] 4) It can provide a preliminary range for fault location for the staff while detecting faults.

[0213] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions that are illustrated and described herein or that are not illustrated and described herein but are understood by those skilled in the art.

[0214] Those skilled in the art will understand that information, signals, and data can be represented using any of a variety of different technologies and techniques. For example, the data, instructions, commands, information, signals, bits (bits), symbols, and chips described above throughout the description can be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or optical particles, or any combination thereof.

[0215] Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability of hardware and software, the various illustrative components, boxes, modules, circuits, and steps are described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and the design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

[0216] The various illustrative logical modules and circuits described in connection with the embodiments disclosed herein can be implemented using a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0217] The steps of a method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read from, and write to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In the alternative, the processor and the storage medium may reside as discrete components in a user terminal.

[0218] As used in this application and the appended claims, unless the context clearly dictates otherwise, the words "a," "an," "the," and / or "said" are not intended to refer to the singular and may include the plural. In general, the terms "comprising" and "including" are used to indicate that the steps and elements listed are included, without excluding the possibility that other steps and elements may be present.

[0219] The above embodiments are provided for those skilled in the art to implement or use the present invention. Those skilled in the art can make various modifications or changes to the above embodiments without departing from the inventive concept of the present invention. Therefore, the protection scope of the present invention is not limited by the above embodiments, but should be the maximum scope that conforms to the innovative features mentioned in the claims.

Claims

1. A distributed process monitoring method for collaborative modeling within and between blocks, characterized in that It includes two stages: offline modeling and online monitoring. The offline modeling stage specifically includes the following steps: Step S1: Decompose the process data to obtain each sub-block, and construct a topological matrix C; Step S2: Model each sub-block using the slow feature analysis method within the sub-block, and extract slow features and retain the projection matrix; Step S3: Calculate statistics based on the extracted slow features respectively, and obtain the corresponding control limits through kernel density estimation; Step S4: Perform canonical correlation analysis modeling between connected sub-blocks according to the topological matrix C, and extract canonical correlation components and retain the projection matrix; Step S5: Generate a residual vector based on the canonical correlation components and calculate statistics, and obtain the corresponding control limits through kernel density estimation; The online monitoring stage specifically includes the following steps: Step S6: For the new sampled data, perform corresponding partitioning according to Step S1 to decompose it into each sub-block; Step S7: Substitute the partitioned sampled data into the slow feature analysis method model within the sub-block and the canonical correlation analysis model between sub-blocks respectively to obtain the corresponding feature components; Step S8: Calculate the statistics of the models corresponding to the slow feature analysis method model and the canonical correlation analysis model respectively; Step S9: For the different detection results of multiple models of the same sub-block, Bayesian inference fusion is used to obtain a comprehensive index ET 2 as the final monitoring statistic, and it is compared with the corresponding control limit to detect the status of the current sub-block; Step S10: Calculate the fault detection index FD of the entire process, and compare it with the corresponding control limit. When it exceeds the control limit, it indicates that there is a fault in the process. Otherwise, the operating state of the process is considered normal.

2. The distributed process monitoring method for collaborative intra-block and inter-block modeling according to claim 1, wherein The said Step S1 further includes: Decompose the process data in combination with the process background and mechanism knowledge to obtain each sub-block, and the corresponding expression is: X = [X1, X2, …, X i …, X B ​ Among them, X i is the i-th sub-block, and B is the number of sub-blocks.

3. The distributed process monitoring method for intra-block and inter-block collaborative modeling according to claim 1, wherein, In the said Step S2, the slow features extracted from each sub-block, the corresponding expression is: s i = W i X i where s i is the slow feature of the i-th sub-block, and W i is the weight matrix corresponding to the i-th sub-block, and X i is the i-th sub-block.

4. The distributed process monitoring method for intra-block and inter-block collaborative modeling according to claim 1, characterized in that The statistics in the said Step S3, the corresponding expression is: Among them, is the statistic corresponding to the i-th sub-block, used to monitor the change of the components representing the essence of the process state in the subspace, s i is the slow feature of the i-th sub-block.

5. The distributed process monitoring method for intra-block and inter-block collaborative modeling according to claim 1, characterized in that, The kernel density estimation The corresponding expression is: where y is the data to be estimated, and y i is the observed value of the process data, n is the number of samples, h is the smoothing parameter, and K(·) is the kernel function.

6. The distributed process monitoring method for intra-block and inter-block collaborative modeling according to claim 1, characterized in that, In the said Step S4, perform canonical correlation analysis modeling between the sub-block and the sub-block with which it has information interaction according to the topological matrix C, and the expression of the obtained canonical correlation components is: Among them, A ij and B ij are the projection matrices of sub-block i and sub-block j, u ij and v ij are the typical components of sub-block i and sub-block j respectively, X i is the i-th sub-block, and X j is the j-th sub-block.

7. The distributed process monitoring method for collaborative intra-block and inter-block modeling according to claim 1, wherein In the said Step S5, the residual vector established based on the extracted canonical correlation components, the corresponding expression is: Among them, r ij is the residual vector between sub-block i and sub-block j, Λ k,ij is the canonical correlation coefficient diagonal matrix of the first k pairs of relevant variables between sub-block i and sub-block j, X i is the i-th sub-block, X j is the j-th sub-block, A ij and B ij are the projection matrices of sub-block i and sub-block j; In the said Step S5, construct statistics based on the residual vector, the corresponding expression is: Among them, is a statistic constructed based on the residual vector between sub-block i and sub-block j, I k is the K-order identity matrix, and n is the number of samples.

8. The distributed process monitoring method for collaborative modeling within and between blocks according to claim 1, wherein The final monitoring statistic ET in step S9 described above 2 , and the corresponding expression is: Among them, is the final monitoring statistic of the i-th sub-block, x new is the online data, is the conditional probability of the abnormal operating state of the process, is the probability of fault occurrence, P(c ij ) = c ij is the coefficient representing the connection relationship between sub-blocks based on the topology matrix C, B is the number of sub-blocks, is under the fault F and the connection relationship c ij condition, the conditional probability of the online data x new , is the fault posterior probability based on the residual statistic.

9. The distributed process monitoring method for intra-block and inter-block collaborative modeling according to claim 8, characterized in that In the step S9 described above, the control limit CL corresponding to the final monitoring statistic of the i-th sub-block i is the prior probability (1-α) of an abnormal operating condition occurring in the process, where α is the prior probability of the process being in a normal operating state.

10. The distributed process monitoring method for intra-block and inter-block collaborative modeling according to claim 9, characterized in that, In the said Step S10, the expression of the fault detection index FD of the entire process is: Among them, the control limit of the fault detection index FD is 1, is the final monitoring statistic of the first sub-block, is the final monitoring statistic of the second sub-block, is the final monitoring statistic of the Bth sub-block.

11. A distributed process monitoring device for collaborative modeling within and between blocks, comprising: A memory for storing instructions executable by a processor; A processor for executing the instructions to implement the method according to any one of claims 1-10.

12. A computer-readable medium having computer instructions stored thereon, wherein when the computer instructions are executed by a processor, the method according to any one of claims 1-10 is executed.

Images