An online monitoring and fault diagnosis method for continuous stirred tank reactor

Abstract

This invention discloses an online monitoring and fault diagnosis method for continuous stirred tank reactors. The method includes training a support vector machine (SVM) model using offline data to obtain a score set for the training set. This score set is then used to train a kernel density estimation model, and a probability threshold is determined based on the data distribution characteristics of the score set. Real-time field data is preprocessed, fed into the SVM to obtain score values, and then the kernel density estimation model is used to obtain predicted probabilities. If the predicted probability value is lower than the probability threshold, a fault is determined to have occurred. The real-time monitored faults are fed into a Shapley additive interpretation model, and after statistically analyzing a sufficient number of fault points, a feature importance ranking map is obtained. Fault diagnosis is then completed by combining expert knowledge. This invention fully considers the complex process and numerous variables of continuous stirred tank reactors, and uses a SVM to establish an online monitoring model for continuous stirred tank reactors.

Description

Technical Field

[0001] This invention relates to the field of online fault monitoring and fault diagnosis technology, and in particular to an online monitoring and fault diagnosis method for continuous stirred tank reactors. Background Technology

[0002] With the continuous development and progress of my country's industry, chemical processes are gradually transitioning from simple single-stage systems with few devices to complex multi-stage systems with many devices. These chemical processes are characterized by strong nonlinearity, dynamism, and strong coupling. They are difficult to accurately describe using mechanistic modeling, and faults become more difficult to monitor and control.

[0003] Continuous stirred tank reactors (CSTRs) are a widely used reactor type in chemical processes. Operating in a continuous mode, they use a stirring device to mix reactants uniformly, enabling the continuous reaction process. They are typically equipped with multiple sensors, such as temperature sensors, flow meters, and pressure sensors, for real-time monitoring of their operating status. However, CSTRs are affected by numerous physical and chemical factors, thus data-driven models that do not require prior physicochemical knowledge, such as Support Vector Machines (SVMs), have been applied. However, the data collected in CSTRs lacks labeling information and exhibits strong imbalance characteristics, making it difficult to model using traditional SVMs. Furthermore, the fault monitoring results obtained by SVMs lack interpretability and have unclear decision-making processes, making it difficult to further infer the fault type and location. To ensure the safety and stability of CSTRs, it is essential to establish an effective online process monitoring and fault diagnosis scheme to address fault phenomena promptly.

[0004] Chapter 1.1 of "Partially Interpretable Soft-Margin Support Vector Machines Based on Nonlinear Prior Knowledge from Data Mining" points out that kernelized support vector machines are black-box models, meaning the series of operations performed by the model during the prediction process are unknown. The aim is to understand what knowledge, expressed in a human-understandable way, the model has learned from the data to arrive at its final decision. Furthermore, model interpretability is a subjective property, characterizing the degree of understanding of the prediction results or model decisions, and currently, it cannot be defined with rigorous mathematical expressions. Interpretability helps in understanding the specific causal relationship between the output and input, and also helps improve the model's reliability.

[0005] Compared with a class of support vector machine methods in "Fault Diagnosis Methods Based on DMVU-OCSVM", using a kernel density estimation model to determine the control threshold can improve the fault detection rate. Furthermore, using a Shapley additive interpretation model can provide an explanation of the causal relationship between input and output, identify fault variables, and complete fault diagnosis. Summary of the Invention

[0006] In order to at least partially solve one of the technical problems existing in the prior art, the purpose of this invention is to provide an online monitoring and fault diagnosis method for continuous stirred reactors.

[0007] The present invention is achieved by at least one of the following technical solutions.

[0008] A method for online monitoring and fault diagnosis of a continuous stirred tank reactor includes the following steps:

[0009] Determine the observation variables and set the sampling interval. Use the observation data collected under normal operating conditions as the training set. Preprocess the training set to obtain the preprocessed training set.

[0010] Based on the features of the training set, determine the kernel function type and corresponding hyperparameters of a class of support vector machines, train a class of support vector machines using the preprocessed training set, and obtain the score set corresponding to the training set;

[0011] The kernel density estimation model is trained using a set of scores, and a probability threshold is determined based on the data distribution characteristics of the set of scores to determine whether a new sample is abnormal.

[0012] To determine whether a continuous stirred tank reactor has malfunctioned: the pre-processed observation sample at that moment is fed into a support vector machine to obtain a score, and then the score is fed into a kernel density estimation model to obtain a predicted probability; if the predicted probability is greater than the probability threshold, the operating condition is determined to be normal and the next sampling is performed; if the predicted probability is less than the probability threshold, a malfunction is determined to have occurred at that observation moment and an alarm is issued.

[0013] The Shapley additive interpretation model interpreter is used to calculate the Shapley value of the fault samples, and statistical methods are used to rank the Shapley values ​​of the fault samples by importance to obtain a feature importance ranking map.

[0014] Based on the feature importance ranking chart, and combined with the characteristics and patterns of failures in continuous stirred tank reactors, expert knowledge is used to identify failure types from the feature importance ranking chart.

[0015] Furthermore, the preprocessing includes:

[0016] Using the standard deviation standardization method, the data matrix observed from the continuous stirred tank reactor is preprocessed according to the following formula:

[0017]

[0018] Where, X∈R m×n It is the original observation data matrix, R m×nLet X be an m x n real matrix, where each row of X represents an observed sample and each column represents an observed variable, μ x and σ x X represents the sample mean and sample standard deviation of each observed variable in X, and X' represents the preprocessed training set.

[0019] Furthermore, the preprocessing of the training set includes:

[0020] Using range normalization, the data matrix observed from the continuous stirred tank reactor is preprocessed according to the following formula:

[0021]

[0022] Where, X∈R m×n It is the original observation data matrix, R m×n Let X be an m x n real matrix, where each row of X represents an observed sample and each column represents an observed variable. Let X' represent the preprocessed training set. Let X(i,j) and X'(i,j) represent the elements at the i-th row and j-th column positions of X and X', respectively. Let X(j) be the element at the i-th row and j-th column of X and X', respectively. max X(j) min Let X represent the maximum and minimum values ​​in the j-th column of X, respectively.

[0023] Furthermore, a class of support vector machines is constructed, whose decision function expression is:

[0024] f(x) = w T φ(x)-ρ

[0025] Where w is a general vector perpendicular to the decision hyperplane, ρ is the intercept term, and φ(x) is the mapping of sample x in the high-dimensional feature space, and sample x is taken from a row vector in the X' matrix.

[0026] To further separate the sample points from the origin, a class of support vector machines is simplified to the following objective equation:

[0027]

[0028] stw T φ(x i )≥ρ-ξ i ,ξ i >0

[0029] Where, ξ i Let v represent the slack variable, v be the upper bound hyperparameter set for the outlier score, m represent the number of row vectors in matrix X', and x represent the slack variable. i This represents a support vector on the decision hyperplane.

[0030] Furthermore, by applying a Lagrange transform to the objective equation and using the kernel function technique for the dot product operation, the decision function becomes:

[0031]

[0032] Where, α i It is the Lagrange multiplier corresponding to each sample, K(x) i (x) is the kernel function, which can be a radial basis function or a linear kernel function:

[0033]

[0034] K(x i ,x)=x i ·x

[0035] Where γ represents the free parameter, x i Let X represent a support vector, and let x represent a row vector in the X' matrix.

[0036] Furthermore, the kernel density estimation model is as follows:

[0037]

[0038] Where m is the total number of samples in the score set, and its value is equal to the number of row vectors in matrix X'; h is the bandwidth; A i Let i be the i-th element in the set of fractions; To predict the probability of a given fractional sample a, the kernel density is estimated using a Gaussian kernel function.

[0039] The model is updated as follows:

[0040]

[0041] Based on the data distribution characteristics of the fraction set, the probability threshold is determined to be α.

[0042] Furthermore, determining whether a continuous stirred tank reactor has malfunctioned includes:

[0043] During the online monitoring phase, the preprocessed observation samples at that moment are fed into a support vector machine to obtain scores, and then these scores are fed into a kernel density estimation model to obtain the predicted probability. Compare with the probability threshold α; if If the operating condition is deemed normal, proceed to the next sampling; if If a fault occurs at that observation time, an alarm will be issued.

[0044] Furthermore, the Shapley value of the fault sample is calculated using the Shapley additive interpretation model interpreter, including:

[0045] An interpreter for the Shapley additive interpretation model is created using the Shapley additive interpretation library. A trained support vector machine model is passed as parameters to the interpreter, which then calculates the Shapley value for each fault sample.

[0046] The following formula calculates the Shapley value for each feature:

[0047]

[0048] in, φ is the j-th feature of the i-th sample. i j Let be the corresponding Shapley value, L be the set of all features in the training set with feature dimension p, and S be the subset of features selected from L with feature dimension |S|. f(S) and f(S) represent the scores of a support vector machine model f with and without feature i added to the feature subset S, respectively; after obtaining the Shapley value corresponding to each feature, the absolute value of the Shapley value corresponding to each feature is taken.

[0049] Furthermore, a feature importance ranking map is obtained using statistical methods, including: counting a certain number of fault samples, ranking and visualizing the Shapley values ​​of these samples, and obtaining a feature importance ranking map containing multiple fault samples.

[0050] Furthermore, based on the feature importance ranking chart, fault types are identified. Combining the characteristics and patterns of fault occurrence in a continuous stirred tank reactor, the different characteristics of the importance ranking of fault features of various sensors are analyzed, and expert knowledge is used to identify fault types from the feature importance ranking chart.

[0051] Another technical solution adopted in this invention is:

[0052] An online monitoring device for a continuous stirred tank reactor includes:

[0053] At least one processor;

[0054] At least one memory for storing at least one program;

[0055] When the at least one program is executed by the at least one processor, the at least one processor implements the method described above.

[0056] Another technical solution adopted in this invention is:

[0057] A computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the method described above.

[0058] Compared with existing technologies, the present invention has at least the following advantages and beneficial effects:

[0059] 1. This invention effectively solves the problems of insufficient fault data and unbalanced label data in continuous stirred tank reactors by learning a classification hyperplane of a support vector machine using data samples from normal operation, and realizes fault monitoring of continuous stirred tank reactors.

[0060] 2. This invention can significantly reduce the false alarm rate and false alarm rate when detecting faults in a continuous stirred tank reactor, thus reducing the workload of continuous stirred tank reactor operators and enabling timely detection of faults, allowing sufficient time for maintenance of chemical plant equipment and saving maintenance costs.

[0061] 3. The Shapley additive interpretation model used in this invention can interpret the scores given by a class of support vector machines. By ranking the importance of variables that cause score anomalies and combining expert knowledge, different types of faults can be identified, thus providing reference information for on-site personnel to make maintenance strategies. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a flowchart of an online monitoring and fault diagnosis method for a continuous stirred tank reactor according to an embodiment of the present invention;

[0064] Figure 2 This is a schematic diagram of a simulated continuous stirred tank reactor used in an embodiment of the present invention;

[0065] Figure 3 This is a feature importance ranking diagram of Embodiment 1 of the present invention;

[0066] Figure 4 This is a feature importance ranking diagram of Embodiment 2 of the present invention;

[0067] Figure 5 This is a feature importance ranking diagram of Embodiment 3 of the present invention;

[0068] Figure 6 This is a feature importance ranking diagram of Embodiment 4 of the present invention. Detailed Implementation

[0069] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0070] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0071] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0072] In the description of this invention, unless otherwise explicitly defined, terms such as "setting," "installing," and "connecting" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0073] Example 1

[0074] like Figure 1As shown, this embodiment provides an online monitoring and fault diagnosis method for a continuous stirred tank reactor. The method first trains a support vector machine (SVM) model (the monitoring model is an online fault monitoring model of the SVM) using data from the continuous stirred tank reactor, i.e., offline data, to obtain a score set for the training set. Then, a kernel density estimation model is trained using this score set, and a probability threshold is determined based on the data distribution characteristics of the score set. The collected real-time data is preprocessed, fed into the SVM to obtain score values, and then the kernel density estimation model is used to obtain the predicted probability. If the predicted probability value is lower than the probability threshold, a fault can be determined. The real-time monitored faults are fed into a Shapley additive interpretation model (the diagnosis model is an online fault diagnosis model of the Shapley additive model). After statistically analyzing a sufficient number of fault points, a feature importance ranking map is obtained, and fault diagnosis is completed by combining expert knowledge. Specifically, the method includes the following steps:

[0075] A1. Determine the observed variables and set the same sampling interval. Use the data matrix collected from the continuous stirred tank reactor under normal operating conditions as the training set, and perform data preprocessing according to the following formula:

[0076]

[0077] Where, X∈R m×n It is the original observation data matrix, R m×n Let X be an m x n real matrix. Each row of X represents an observation sample, and each column represents an observation variable, including input material temperature, output material temperature, etc. x and σ x X represents the sample mean and sample standard deviation of each observed variable in X, and X' represents the preprocessed training set.

[0078] A2. Set the hyperparameter v of a class of support vector machines, and determine the kernel function as the radial basis kernel function.

[0079] A3. Train the kernel density estimation model using a fraction set, and determine the probability threshold α based on the data distribution characteristics of the fraction set;

[0080] A4. Collect real-time observation data as test samples. After preprocessing the test set, use a support vector machine to predict scores, and then feed the scores into a kernel density estimation model to obtain the predicted probabilities.

[0081] A5. Determine if a continuous stirred tank reactor has malfunctioned. (Predict the probability.) Compared with a pre-determined probability threshold α, if If the operating condition is deemed normal, proceed to the next sampling; if If a fault occurs at that observation time, an alarm will be issued.

[0082] A6. Create a Shapley additive interpreter, pass a trained support vector machine model as a parameter to the interpreter, and then use the interpreter to calculate the Shapley value of the detected fault samples.

[0083] A7. Sort the Shapley values ​​of each fault sample in descending order, and then rank the importance of each feature in each fault sample to form a feature importance ranking chart containing multiple samples.

[0084] A8. Based on the characteristics and patterns of faults occurring in a continuous stirred tank reactor, analyze the different characteristics of sensor drift faults and thermal conductivity faults, and use expert knowledge to identify fault types from the feature importance ranking chart.

[0085] In this embodiment, data from a simulated continuous stirred reactor are used for verification, such as... Figure 2 As shown in Table 1, the dataset is provided by a MATLAB simulation of a continuous stirred tank reactor system. It contains 2400 samples with a sampling interval of one minute. The first 1200 samples are used as the training set, and the last 1200 samples are used as the test set. Faults are introduced starting from the 200th sample in the test set. There are a total of 7 observed variables, and the meaning of each observed variable is shown in Table 1.

[0086] Table 1 Experimental Data Observation Variables

[0087]

[0088]

[0089] like Figure 1 As shown, the general process of the above-mentioned continuous stirred tank reactor monitoring and fault diagnosis method is as follows: Simulate different types of faults and generate observation data, which is then divided into two parts: the first 1200 minutes of samples serve as the training set for model building, and the remaining 1200 minutes serve as the test set. Different types of faults are introduced at the 200th minute to verify the effectiveness of the proposed method. The specific steps are as follows:

[0090] B1. Preprocess the training data using the standard deviation standardization method;

[0091] B2. Set the hyperparameter v = 0.01 for a class of support vector machines, and determine the radial basis function kernel function. Train a class of support vector machines using the preprocessed dataset X', and set γ to 0.1;

[0092] B3. Train the kernel density estimation model using a fraction set, and determine the probability threshold α as 0.99 based on the data distribution characteristics of the fraction set;

[0093] B4. Collect real-time observation data as test samples. After preprocessing the test set, use a support vector machine to predict scores, and then feed the scores into a kernel density estimation model to obtain the predicted probabilities.

[0094] B5. Determine if a continuous stirred tank reactor has malfunctioned, and predict the probability. Compared with a pre-determined probability threshold α, if If the operating condition is deemed normal, proceed to the next sampling; if If a fault occurs at that observation time, an alarm will be issued.

[0095] B6. Create a Shapley additive interpreter, pass the trained support vector machine model as a parameter to the interpreter, and then use the interpreter to calculate the Shapley value of the detected fault samples.

[0096] B7. Sort the Shapley values ​​of each fault sample in descending order, and then calculate the importance ranking of each feature in each fault sample to form a feature importance ranking chart containing multiple samples.

[0097] B8. Based on the characteristics and patterns of failures in a continuous stirred tank reactor, analyze the resulting feature importance ranking chart, and identify the failure type according to the order of the top-ranked feature variables.

[0098] Experimental results show that this scheme has a good monitoring effect on simulated Ci sensor drift faults. During the online monitoring phase, a Ci sensor fault was introduced into the continuous stirred reactor at the 200-minute mark. The trained model was able to quickly detect the fault, with a recall rate of 98.5% and a false alarm rate of 4.5%. Furthermore, after detecting 100 fault points, a feature importance ranking chart was obtained through statistical analysis, as shown below. Figure 3 As shown in the chart, the Ci variable ranks first and contributes the most to this type of fault, so it can be inferred that the fault is a Ci sensor fault.

[0099] Example 2

[0100] Using different experimental data than in Example 1, a Qc sensor fault was simulated, while the parameters set during the modeling process were the same as in Example 1.

[0101] Experimental results show that this scheme has a good monitoring effect on simulated Qc sensor drift faults. During the online monitoring phase, a Qc sensor fault was introduced into the continuous stirred reactor at the 200-minute mark. The trained model was able to quickly detect the fault, with a fault recall rate of 97.4% and a false alarm rate of 1.5%. Furthermore, after detecting 100 fault points, a feature importance ranking chart was obtained through statistical analysis, as shown below. Figure 4As shown in the chart, the Qc variable ranks first and contributes the most to this type of fault, so it can be inferred that the fault is a Qc sensor fault.

[0102] Example 3

[0103] Using different experimental data than in Example 1, a Tc sensor fault was simulated, while the parameters set during the modeling process were the same as in Example 1.

[0104] Experimental results show that this scheme has a good monitoring effect on simulated Tc sensor drift faults. During the online monitoring phase, a Tc sensor fault was introduced into the continuous stirred reactor at the 200-minute mark. The trained model was able to quickly detect the fault, with a fault recall rate of 99.3% and a false alarm rate of 0. Furthermore, after obtaining 100 fault points, a feature importance ranking chart was obtained through statistical analysis, as shown below. Figure 5 As shown in the chart, the Tc variable ranks first and contributes the most to this type of fault, so it can be inferred that the fault is a Tc sensor fault.

[0105] Example 4

[0106] Using different experimental data than in Example 1, a thermal conductivity coefficient fault was simulated. This type of fault differs from sensor drift faults and is a process fault. The parameters set during the modeling process were the same as in Example 1.

[0107] Experimental results show that this scheme has a good monitoring effect on simulated thermal conductivity coefficient faults. During the online monitoring phase, a thermal conductivity coefficient fault was introduced into the continuous stirred reactor at the 200-minute mark. The trained model was able to quickly detect the fault, with a fault recall rate of 97.7% and a false alarm rate of 3%. Furthermore, after obtaining 100 fault points, a feature importance ranking chart was obtained through statistical analysis, as shown below. Figure 6 As shown in the figure, the statistical chart shows that Tc is ranked first and Qc is ranked second. This is significantly different from the feature ranking chart of Tc sensor fault in Example 3. Considering that thermal conductivity coefficient faults are process faults, it can be inferred that the fault is a thermal conductivity coefficient fault.

[0108] The above embodiments demonstrate that the continuous stirred tank reactor monitoring and fault diagnosis method developed in this invention offers more satisfactory performance compared to traditional fault monitoring models. This invention fully considers the complex process and numerous variables inherent in continuous stirred tank reactors, and establishes an online monitoring model for these reactors using a class of support vector machines. However, the monitoring results obtained from the model lack interpretability. This invention uses Shapley additive interpretation combined with expert knowledge to further infer the fault type.

[0109] This embodiment provides an online monitoring device for a continuous stirred tank reactor, comprising:

[0110] At least one processor;

[0111] At least one memory for storing at least one program;

[0112] When the at least one program is executed by the at least one processor, the at least one processor implements Figure 1 The method shown.

[0113] This embodiment provides an online monitoring device for a continuous stirred tank reactor. It can execute the online monitoring method for a continuous stirred tank reactor provided in the method embodiment of the present invention, and can execute any combination of the implementation steps of the method embodiment, possessing the corresponding functions and beneficial effects of the method.

[0114] This application also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform... Figure 1 The method shown.

[0115] This embodiment also provides a storage medium storing instructions or programs that can execute the online monitoring method for a continuous stirred tank reactor provided in the method embodiment of the present invention. When the instructions or programs are run, any combination of implementation steps of the method embodiment can be executed, and the method has the corresponding functions and beneficial effects.

[0116] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0117] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0118] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0119] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0120] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0121] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0122] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0123] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0124] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for online monitoring and fault diagnosis of a continuous stirred tank reactor, characterized in that, Includes the following steps: Determine the observation variables and set the sampling interval. Use the observation data collected under normal operating conditions as the training set. Preprocess the training set to obtain the preprocessed training set. Based on the features of the training set, determine the kernel function type and corresponding hyperparameters of a class of support vector machines, train a class of support vector machines using the preprocessed training set, and obtain the score set corresponding to the training set; The kernel density estimation model is trained using a set of scores, and a probability threshold is determined based on the data distribution characteristics of the set of scores to determine whether a new sample is abnormal. To determine whether a continuous stirred tank reactor has malfunctioned: the pre-processed observation sample at the observation time is fed into a support vector machine to obtain a score value, and then the score value is fed into a kernel density estimation model to obtain a predicted probability; if the predicted probability is greater than the probability threshold, the operating condition is determined to be normal and the next sampling is performed; if the predicted probability is less than the probability threshold, the malfunction at the observation time is determined to have occurred and an alarm is issued. The Shapley additive interpretation model interpreter is used to calculate the Shapley value of the fault samples, and statistical methods are used to rank the Shapley values ​​of the fault samples by importance to obtain a feature importance ranking map. The Shapley values ​​for fault samples are calculated using the Shapley additive interpretation model interpreter, including: An interpreter for the Shapley additive interpretation model is created using the Shapley additive interpretation library. A pre-trained support vector machine model is passed as parameters to the interpreter, which then calculates the Shapley value for each fault sample. The interpreter uses the following formula to calculate the Shapley value for each feature: in, For the j-th feature of the i-th sample, Let be the corresponding Shapley value, L be the set of all features in the training set with feature dimension p, and S be the subset of features selected from L with feature dimension |S|. f(S) and f(S) represent the scores of a support vector machine model f with and without feature i added to the feature subset S, respectively; after obtaining the Shapley value corresponding to each feature, the absolute value of the Shapley value corresponding to each feature is taken. Based on the feature importance ranking chart, and combined with the characteristics and patterns of failures in a continuous stirred tank reactor, expert knowledge is used to identify failure types from the feature importance ranking chart. Statistical methods are used to obtain the feature importance ranking chart, including: counting a certain number of failure samples, ranking and visualizing the Shapley values ​​of these samples, and obtaining a feature importance ranking chart containing multiple failure samples.

2. The method for online monitoring and fault diagnosis of a continuous stirred tank reactor according to claim 1, characterized in that, The preprocessing includes: Using the standard deviation standardization method, the data matrix observed from the continuous stirred tank reactor is preprocessed according to the following formula: Where, X∈R m×n It is the original observation data matrix, R m×n Let X be an m x n real matrix, where each row of X represents an observed sample and each column represents an observed variable, μ x and σ x X represents the sample mean and sample standard deviation of each observed variable in X, and X' represents the preprocessed training set.

3. The method for online monitoring and fault diagnosis of a continuous stirred tank reactor according to claim 1, characterized in that, The preprocessing of the training set includes: Using range normalization, the data matrix observed from the continuous stirred tank reactor is preprocessed according to the following formula: Where, X∈R m×n It is the original observation data matrix, R m×n Let X be an m x n real matrix, where each row of X represents an observed sample and each column represents an observed variable. Let X' represent the preprocessed training set, and X(j) be the training set. max X(j) min Let X(i,j) and X'(i,j) represent the maximum and minimum values ​​in the j-th column of X, respectively. Let X(i,j) and X'(i,j) represent the elements in the i-th row and j-th column of the original observation data matrix X and the preprocessed training set data matrix X', respectively.

4. The method for online monitoring and fault diagnosis of a continuous stirred tank reactor according to claim 1, characterized in that, Construct a class of support vector machines whose decision function expression is: f(x)=w T φ(x)-ρ Where w is a general vector perpendicular to the decision hyperplane, ρ is the intercept term, and φ(x) is the mapping of sample x in the high-dimensional feature space, and sample x is taken from a row vector in the X' matrix. To further separate the sample points from the origin, a class of support vector machines is simplified to the following objective equation: stw T φ(x i )≥ρ-ξ i ,x i >0 Where, ξ i Let v represent the slack variable, v be the upper bound hyperparameter set for the outlier score, m represent the number of row vectors in matrix X', and x represent the slack variable. i This represents a support vector on the decision hyperplane.

5. The method for online monitoring and fault diagnosis of a continuous stirred tank reactor according to claim 4, characterized in that, Applying a Lagrange transform to the objective equation and using the kernel function technique for the dot product operation, the decision function becomes: Where, α i It is the Lagrange multiplier corresponding to each sample, K(x) i (x) is the kernel function, which can be a radial basis function or a linear kernel function: K'(x i ,x)=x i ·x Where γ represents the free parameter, x i Let X represent a support vector on the decision hyperplane, and let x represent a row vector in the X' matrix.

6. The method for online monitoring and fault diagnosis of a continuous stirred tank reactor according to claim 1, characterized in that, The kernel density estimation model is as follows: Where m is the total number of samples in the score set, and its value is equal to the number of row vectors in matrix X'; h is the bandwidth; A i Let i be the i-th element in the set of fractions; For the predicted probability of a given fractional sample a, the kernel density estimation model is updated using the Gaussian kernel function as follows: Based on the data distribution characteristics of the fraction set, the probability threshold is determined to be α.

7. The method for online monitoring and fault diagnosis of a continuous stirred tank reactor according to claim 1, characterized in that, The determination of whether a continuous stirred tank reactor has malfunctioned includes: During the online monitoring phase, the preprocessed observation samples at that moment are fed into a support vector machine to obtain scores, and then these scores are fed into a kernel density estimation model to obtain the predicted probability. Compare with the probability threshold α; if If the operating condition is deemed normal, proceed to the next sampling; if If a fault occurs at that observation time, an alarm will be issued.

8. The method for online monitoring and fault diagnosis of a continuous stirred tank reactor according to claim 1, characterized in that, Based on the feature importance ranking chart, fault types are identified. Combining the characteristics and patterns of fault occurrence in a continuous stirred tank reactor, the different characteristics of the feature importance ranking of various sensors are analyzed, and expert knowledge is used to identify fault types from the feature importance ranking chart.

Images