Tin chemical process anomaly detection method, device and system
By constructing a state space model and using PPFA algorithm, the accuracy and reliability of abnormal detection in the tin chemical process are solved, the intelligent control level of the tin chemical production process is improved, and the digital transformation of the industry is promoted.
Patent Information
- Application Number
- CN202411143570.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-22
AI Technical Summary
The existing technology cannot effectively use the production data of the tin chemical process for abnormal detection, resulting in inaccurate and unreliable detection results, especially in the face of multivariable, nonlinear and strong coupling production processes, especially when product switching, Monitoring and control become more complex.
By acquiring sensor data, a state space model is constructed to reflect the mapping relationship between the data and the dynamic latent variable characteristics. The dynamic latent variable characteristics are extracted using the probability predictable feature analysis (PPFA) algorithm, statistical limits are calculated and compared, and tin is determined. Whether abnormalities occurred in the chemical process.
It significantly improves the accuracy and reliability of abnormal detection of tin chemical process, improves the level of intelligent control, and promotes the digital transformation of the industry.
Smart Images

Figure CN119004332A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the technical field of tin chemical process detection, and in particular to a method, device and system for detecting abnormalities in a tin chemical process. Background Art
[0002] As one of the most important tin chemical products, methyltin plays a pivotal role and is widely used in integrated circuit packaging, ship coating, automotive electrophoretic paint catalyst, liquid crystal display, pharmaceutical intermediates, electronics, polyurethane catalyst, building materials, engineering plastics, flame retardants, food additives, textiles and other fields. Its production process has the characteristics of large hysteresis, multivariable, nonlinear, strong coupling, etc., which puts high demands on the safety and smooth operation of the reaction process. In addition, when producing different products, tin chemical reactions need to face the random switching between products, which increases the complexity of the production process and poses higher challenges to the monitoring and control of the production process.
[0003] With the rapid development and application of distributed control systems (DCS), a large number of sensors are installed in the tin chemical process to measure production data, which provides a data basis for monitoring the tin chemical process. However, there is currently no better solution for how to use these production data to detect anomalies in the tin chemical process and ensure that the detection results are reliable and accurate. Summary of the invention
[0004] In order to solve the problem that the prior art cannot effectively use production data to detect anomalies in tin chemical processes and ensure that the detection results are reliable and accurate, the embodiments of the present application provide a method, device and system for detecting anomalies in tin chemical processes.
[0005] According to a first aspect of an embodiment of the present application, a method for detecting abnormality in a tin chemical process is provided, comprising:
[0006] Obtain production data collected by sensors in the tin chemical process;
[0007] Preprocessing the production data to obtain a target data set corresponding to the production data;
[0008] Building a state space model based on the target data set, wherein the state space model has a probability structure in the form of a state space, and is used to reflect the mapping relationship between the data of the target data set and the dynamic latent variable characteristics;
[0009] Extracting dynamic latent variable features corresponding to the training data set through the state space model, and calculating statistical limits according to the dynamic latent variable features corresponding to the training data set;
[0010] Extracting dynamic latent variable features corresponding to the test data set through the state space model, and calculating statistics according to the dynamic latent variable features corresponding to the test data set;
[0011] The statistical value is compared with the statistical limit value, and whether an abnormality occurs in the tin chemical process is determined based on the comparison result.
[0012] Optionally, constructing a state space model based on the target data set includes:
[0013] Constructing an initial state space model mapping the data of the target data set to dynamic latent variable features through a probabilistic predictable feature analysis (PPFA) algorithm;
[0014] The parameters of the initial state space model are estimated and optimized to obtain the state space model.
[0015] Optionally, estimating and optimizing parameters of the initial state space model includes:
[0016] The parameters of the initial state space model are estimated and optimized by using the maximum likelihood method and / or the expectation maximization algorithm.
[0017] Optionally, the preprocessing the production data to obtain a target data set corresponding to the production data includes:
[0018] Obtaining an original data set corresponding to the production data;
[0019] Dividing the original data set into a plurality of sub-data sets;
[0020] Calculating the Hausdorff distance between the plurality of sub-datasets, wherein the Hausdorff distance reflects the similarity between the sub-datasets;
[0021] Eliminating abnormal data in the sub-dataset according to the Hausdorff distance;
[0022] The plurality of sub-data sets from which abnormal data have been eliminated are integrated to obtain a target data set corresponding to the production data.
[0023] Optionally, the obtaining of an original data set corresponding to the production data includes:
[0024] According to the time information corresponding to the production data, the production data is divided into initial stage data, severe stage data and stable stage data;
[0025] Performing noise filtering on the data in the initial stage to obtain a first data set;
[0026] Extracting change features from the severe stage data to obtain a second data set;
[0027] Using the stable phase data as the third data set;
[0028] The first data set, the second data set and the third data set are combined to obtain the original data set.
[0029] Optionally, dividing the original data set into a plurality of sub-data sets includes:
[0030] A sliding window operation is performed on the original data set with a preset window length and a preset sliding step length to obtain a plurality of sub-data sets.
[0031] Optionally, calculating the Hausdorff distance between the plurality of sub-datasets includes:
[0032] The Hausdorff distance between the probability density functions of the two adjacent sub-datasets in the window is calculated.
[0033] Optionally, the production data includes continuous process data and discrete operation data;
[0034] The continuous process data includes at least one of the following: temperature, pressure, liquid level;
[0035] The discrete operation data includes at least one of the following: a heating switch, a cooling water switch, a stirring motor, and an air intake valve.
[0036] According to a second aspect of an embodiment of the present application, a tin chemical process abnormality detection device is provided, comprising:
[0037] An acquisition module is used to acquire production data collected by sensors in the tin chemical process;
[0038] A preprocessing module, used to preprocess the production data to obtain a target data set corresponding to the production data;
[0039] A construction module, used to construct a state space model based on the target data set, wherein the state space model has a probability structure in the form of a state space, and is used to reflect the mapping relationship between the data of the target data set and the dynamic latent variable characteristics;
[0040] A first calculation module, configured to extract dynamic latent variable features corresponding to the training data set through the state space model, and calculate statistical limits according to the dynamic latent variable features corresponding to the training data set;
[0041] A second calculation module is used to extract dynamic latent variable features corresponding to the test data set through the state space model, and calculate statistics according to the dynamic latent variable features corresponding to the test data set;
[0042] The comparison module is used to compare the statistical value with the statistical limit value, and determine whether an abnormality occurs in the tin chemical process according to the comparison result.
[0043] According to a third aspect of an embodiment of the present application, a tin chemical process abnormality detection system is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the tin chemical process abnormality detection method described in any one of the first aspects when executing the program.
[0044] The technical solution provided by this application may have the following beneficial effects:
[0045] The embodiment of the present application provides a method for detecting anomalies in a tin chemical process. Since a state space model is constructed using production data, the state space model has a probability structure in the form of a state space, which is used to reflect the mapping relationship between data and dynamic latent variable characteristics. Therefore, under the assumption that random noise obeys a Gaussian distribution, the solution provided by the embodiment of the present application is superior to deterministic methods (such as dynamic internal principal component analysis (DiPCA) and predictable feature analysis (PFA)). Through the probability structure in the form of a state space, the dynamic characteristics of the variables, the influence of random noise and outliers are fully considered, which significantly improves the interpretability and practicality of the model, thereby improving the accuracy and reliability of the anomaly detection results of the tin chemical process. When used in tin chemical processes, this innovative solution will improve the level of intelligent control of tin chemical production processes and promote the digital transformation of the industry.
[0046] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory and do not limit the embodiments of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] The drawings herein are incorporated into the specification and constitute a part of the embodiments of the present application, illustrate embodiments consistent with the embodiments of the present application, and together with the description are used to explain the principles of the embodiments of the present application.
[0048] Figure 1 This is a flow chart of a method for detecting abnormalities in a tin chemical process provided in one embodiment of the present application.
[0049] Figure 2 It is a structural block diagram of a tin chemical process anomaly detection device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0050] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the appended claims.
[0051] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0052] It should be understood that, although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present application, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the embodiments of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at..." or "when..." or "in response to determination".
[0053] As one of the most important tin chemical products, methyltin plays a pivotal role and is widely used in various fields. Its production process has the characteristics of large hysteresis, multivariable, nonlinear, strong coupling, etc., which puts high demands on the safety and smooth operation of the reaction process. In addition, when producing different products, tin chemical reactions need to face the random switching between products, which increases the complexity of the production process and poses higher challenges to the monitoring and control of the production process.
[0054] With the rapid development and application of distributed control systems (DCS), a large number of sensors are installed in the tin chemical process to measure various process variables. Multivariate statistical process monitoring methods such as principal component analysis (PCA), partial least squares (PLS) and canonical correlation analysis (CCA) are widely used. However, traditional principal component analysis methods fail to consider time information and usually lead to poor performance. In order to deal with the dynamic changes in data samples, dynamic principal component analysis (DPCA) is proposed. Although DPCA considers some dynamic characteristics, its internal structure is still static and cannot provide an explicit relationship between latent variables and observed values, which limits the interpretability of the model.
[0055] In order to solve the above problems, the embodiment of the present application provides a method for detecting abnormalities in a tin chemical process. It can be understood that the method provided in the embodiment of the present application can also be applied to chemical processes of other metals.
[0056] You can refer to Figure 1 , Figure 1 : is a flow chart of a method for detecting abnormality in a tin chemical process provided by an embodiment of the present application, the method comprising the following steps:
[0057] S101: Acquire production data collected by sensors in the tin chemical process.
[0058] With the promotion and application of distributed control systems, a large number of sensors are deployed in the tin chemical process, and production data in the tin chemical process can be collected through the sensors.
[0059] S102: Preprocess the production data to obtain a target data set corresponding to the production data.
[0060] The obtained production data needs to be processed before being used for analysis. These processes can be summarized as preprocessing of the production data. After preprocessing, the target data set corresponding to the production data can be obtained. The target data set is the data set required to build the model.
[0061] S103: Construct a state space model based on the target data set.
[0062] Among them, the state space model has a probabilistic structure in the form of state space, which is used to reflect the mapping relationship between the data of the target data set and the characteristics of the dynamic latent variables. Through the probabilistic structure in the form of state space, the dynamic characteristics of the variables, the influence of random noise and outliers are fully considered, which significantly improves the interpretability and practicality of the model.
[0063] S104: extracting dynamic latent variable features corresponding to the training data set through the state space model, and calculating statistical limits according to the dynamic latent variable features corresponding to the training data set.
[0064] S105: extracting dynamic latent variable features corresponding to the test data set through the state space model, and calculating statistics according to the dynamic latent variable features corresponding to the test data set.
[0065] S106: Compare the statistical value with the statistical limit value, and determine whether an abnormality occurs in the tin chemical process according to the comparison result.
[0066] The statistical value is the test value corresponding to the test data in the online test phase, and the statistical limit is the standard value corresponding to the training data in the offline modeling phase. If the statistical value exceeds the statistical limit, the system determines that the tin chemical process is abnormal at this time, and if the statistical value does not exceed the statistical limit, the system determines that the tin chemical process is in a normal state at this time. In one embodiment, if it is determined that an abnormality occurs, an alarm can be issued for intervention.
[0067] The embodiment of the present application provides a method for detecting anomalies in a tin chemical process. Since a state space model is constructed using production data, the state space model has a probability structure in the form of a state space, which is used to reflect the mapping relationship between data and dynamic latent variable characteristics. Therefore, under the assumption that random noise obeys a Gaussian distribution, the solution provided by the embodiment of the present application is superior to deterministic methods (such as dynamic internal principal component analysis (DiPCA) and predictable feature analysis (PFA)). Through the probability structure in the form of a state space, the dynamic characteristics of the variables, the influence of random noise and outliers are fully considered, which significantly improves the interpretability and practicality of the model, thereby improving the accuracy and reliability of the anomaly detection results of the tin chemical process. When used in tin chemical processes, this innovative solution will improve the level of intelligent control of tin chemical production processes and promote the digital transformation of the industry.
[0068] In one embodiment, the collected production data may include process data and operation data. The process data may be, for example, temperature, pressure, liquid level, etc. Such process data is continuous data and may be referred to as continuous process data. The operation data may be operation data of components such as heating switches, cooling water switches, stirring motors, and air intake valves. Such operation data is discrete data and may be referred to as discrete operation data.
[0069] Since production data contains both continuous process data and discrete manipulation data, the existence of discrete data will lead to abnormal mutation points in the data, which seriously affects the accuracy of anomaly detection in the tin chemical process. In order to solve this problem, the production data can be preprocessed to obtain a target data set that does not contain abnormal data.
[0070] In one embodiment, preprocessing the production data may include multiple processes. In one process, the production data collected by the sensor may be selected, for example, the production data may be selected according to a preset time interval (5 seconds, 10 seconds, etc.).
[0071] In one process, it can include classifying the production data and processing different types of data differently. In one embodiment, the production data can be divided into initial stage data, intense stage data and stable stage data according to the time information corresponding to the production data. Here, the time information of the production data can be the timestamp of the production data, and the production data can be divided into initial stage data, intense stage data and stable stage data based on the timestamp. The initial stage refers to the initial time period when the tin chemical process starts. The initial stage data can be noise filtered and baseline adjusted to ensure that the data reflects the true state of the system rather than the interference at startup, and the first data set is obtained. The intense stage is the time period when the continuous process data (such as temperature, pressure, reaction rate, etc.) changes significantly. This stage focuses on the change trend and rate of the continuous process data. Therefore, the change feature extraction of the intense stage data can be performed, and the change feature can reflect the characteristics of the data change trend or rate. For example, the method of differential, Fourier transform, etc. can be used to analyze the change law of the data to obtain the second data set. In the stable stage, the continuous process data tends to be stable and no longer has significant fluctuations, so the original data can be retained, and the stable stage data is used as the third data set. Finally, the first data set, the second data set and the third data set are merged to obtain the original data set to ensure the completeness of the data.
[0072] In one processing, the original data set may also be normalized. Normalizing the original data set can avoid differences between the data. The specific expression is as follows:
[0073]
[0074] in, represents the original data set, Represents the original data set The mean of Represents the original data set The standard deviation of .
[0075] After the normalization process is completed, in one embodiment, the normalized original data set can be subjected to the elimination of abnormal data. Although the probability structure in the form of a state space can describe the local process dynamics in a more compact and clearer way, and fully considers the dynamic characteristics of variables in the production process, the influence of random noise and outliers, in actual scenarios, industrial processes usually work under various operating conditions. When there are changes and interferences in the system, it is difficult for the local model to cover all operating conditions, resulting in invalid and unexpected results. In addition, the existence of discrete data will cause abnormal mutation points in the data, which seriously affects the accuracy of anomaly detection in the tin chemical process. Therefore, it is necessary to propose abnormal data for the original data set.
[0076] Specifically, the original data set may be divided into a plurality of sub-data sets. There are many ways to divide the sub-data sets. In one embodiment, a sliding window operation may be performed on the original data set with a preset window length and a preset sliding step length to obtain a plurality of sub-data sets. For example, the preset window length may be N, and the preset sliding step length may be 1.
[0077] After dividing and obtaining multiple sub-datasets, the Hausdorff distance between the multiple sub-datasets can be calculated, and the abnormal data in the sub-datasets can be eliminated according to the Hausdorff distance between the sub-datasets. The Hausdorff distance can reflect the similarity between the two sub-datasets. It can be understood that the larger the Hausdorff distance between the two sub-datasets, the lower the similarity between the two sub-datasets, and vice versa. In one embodiment, the Hausdorff distance between the probability density functions of the two sub-datasets adjacent to the window can be calculated. Since the different data between the two sub-datasets adjacent to the window is the last data in the rear window when the sliding step length is 1, if the Hausdorff distance corresponding to the two sub-datasets adjacent to the window is greater than the distance threshold, it can be determined that the last data of the sub-dataset in the rear window is abnormal data, and the abnormal data is eliminated. The selection of the distance threshold can be adjusted according to the actual data situation and system requirements. The appropriate threshold can be determined by historical data analysis and experiments to ensure the accuracy of outlier detection and the stability of the system.
[0078] The specific calculation process is as follows:
[0079] Step A1, define distance: define A={a1,…,a m} and B={b1,…,b n} is a data set, each set contains a finite number of data points. Let d(a, b) be the Euclidean distance between two data points, then the distance from data point a to data set B is:
[0080]
[0081] Step B1, direct distance: The direct distance from dataset A to dataset B is:
[0082]
[0083] Similarly, the direct distance from dataset B to dataset A is:
[0084]
[0085] Step C1, Hausdorff distance definition: The Hausdorff distance between two data sets is defined as follows:
[0086] HD(A,B)=max(D(A,B),D(B,A))
[0087] Step D1, outlier removal: The difference between the two data sets is evaluated using the Hausdorff distance probability density function between the two data sets. The pdf represents the shape of the data distribution, as shown in the following formula:
[0088] hd(i)=HD[pdf(xi(t)),pdf(xi+1(t))]
[0089] Calculate the Hausdorff distance between the probability density function of set A and the probability density function of set B, and use HD as the basis to screen virtual triggering points and outliers. Locate the local maximum of the xith Hausdorff distance, move one Hausdorff distance behind it and compare them to get the HD value (Hausdorff distance) between the two. If the HD value reaches the distance threshold D, the outlier is eliminated.
[0090] In the above embodiment, the similarity measurement method based on Hausdorff distance is adopted, which can effectively eliminate virtual triggering points and outliers in the data, solve the mutation problem caused by discrete manipulation of data, and provide a high-quality data foundation for subsequent dynamic feature extraction and anomaly detection.
[0091] After the abnormal data is extracted, multiple sub-datasets from which the abnormal data have been eliminated can be integrated to obtain a target data set corresponding to the production data.
[0092] In one embodiment, a state space model is constructed based on a target data set. Specifically, a state space initial model that maps data of the target data set to dynamic latent variable features can be constructed by a probabilistic predictable feature analysis (PPFA) algorithm, and then the parameters of the state space initial model are estimated and optimized to obtain the state space model. In one embodiment, estimating and optimizing the parameters of the state space initial model may include: estimating and optimizing the parameters of the state space initial model by a maximum likelihood method and / or an expectation maximization algorithm.
[0093] As an example, when constructing a state space model, the following process may be included:
[0094] Step A2, construct a state space model: construct a map of the kth data in the dataset X to the dynamic latent variable feature t k The PPFA generates the state space form and derives t k The predicted value of The specific expression is as follows:
[0095]
[0096] in represents the emission matrix; Where (1≤j≤s) represents the transfer matrix; e k~Ν(0,Γ), where Satisfies Gaussian distribution; ε k ~N(0,∑), represents the measurement noise; E[t k ]=0, I r belongs to the r×r identity matrix.
[0097] Bj The relationship between and Γ is: yes The covariance of , 1≤i≤r represents the i-th dynamic latent variable feature at time k.
[0098] Step B2, parameter estimation: The parameters of PPFA that need to be estimated are summarized as Θ = {B j (1≤i≤s),H,Γ,∑}, here we can use the maximum likelihood method to optimize the parameters of the probability model. Given the time series X, the complete data of the dynamic system:
[0099]
[0100] in Represents the latent variable matrix. For each sample x k , based on the property of conditional independence, the corresponding high-order linear Markov Gaussian dynamic system It is expressed as:
[0101]
[0102] in p(t1)=p(t2)=…=p(t s )=N(0,I r ) from t k to x k The probability distribution of the mapping is p(x k |t k )=p(x k |t k ,H,Σ)=N(Ht k ,Σ).
[0103] Step C2, use the expectation maximization EM algorithm to estimate Θ: The EM algorithm is mainly divided into two steps of iteration, E step and M step. The E step is to initialize the Θ value in the model, and use the implicit function expectation value obtained by the iteration parameter Θ of step B2 to replace the implicit function to obtain the unknown Q function. The M step is to update the Θ value of the obtained Q function:
[0104] Θ new = argmax(Θ,Θ old )
[0105] The above embodiment extends DiPCA and PFA into a probabilistic structure with a state space form, namely, the Probabilistic Predictable Feature Analysis (PPFA) algorithm. The PPFA algorithm is used to extract the dynamic latent variable characteristics of the tin chemical process, capture the potential dynamic characteristics in the data, optimize parameter estimation, enhance the model's ability to process complex industrial process data, and improve the reliability and accuracy of the model.
[0106] After the state space model is constructed, the dynamic latent variable features of the data can be extracted, for example, the dynamic latent variable features can be extracted for the training data set, and the dynamic latent variable features can also be extracted for the test data set. In one embodiment, the training data set and the test data set can be data different from the target data set. In one embodiment, the training data set and the test data set can also be obtained by dividing the data of the target data set, for example, dividing a part of the data in the target data set as the training data set and another part of the data in the target data set as the test data set according to a specific ratio (such as 7:3).
[0107] When calculating the statistical limit value according to the dynamic latent variable characteristics corresponding to the training data set and calculating the statistical quantity according to the dynamic latent variable characteristics corresponding to the test data set, in one embodiment, the calculation can be performed based on the established statistical method. The specific calculation formula is as follows:
[0108]
[0109] where x new is the dynamic latent variable feature extracted from the training dataset, Λ e is the covariance of N training samples:
[0110]
[0111] where y new is the dynamic latent variable feature extracted from the test dataset, Λ v is the covariance of the M test samples:
[0112] The embodiment of the present application realizes the use of only local samples as training sets, fully utilizing the completeness of these samples, and quickly and accurately realizing abnormal detection of tin chemical processes. Through the optimized training sample selection and processing method, the detection performance of the model can be improved, ensuring timely warning and intervention of abnormal situations in tin chemical processes in practical applications.
[0113] The embodiment of the present application provides a method for detecting anomalies in a tin chemical process. Since a state space model is constructed using production data, the state space model has a probability structure in the form of a state space, which is used to reflect the mapping relationship between data and dynamic latent variable characteristics. Therefore, under the assumption that random noise obeys a Gaussian distribution, the solution provided by the embodiment of the present application is superior to deterministic methods (such as dynamic internal principal component analysis (DiPCA) and predictable feature analysis (PFA)). Through the probability structure in the form of a state space, the dynamic characteristics of the variables, the influence of random noise and outliers are fully considered, which significantly improves the interpretability and practicality of the model, thereby improving the accuracy and reliability of the anomaly detection results of the tin chemical process. When used in tin chemical processes, this innovative solution will improve the level of intelligent control of tin chemical production processes and promote the digital transformation of the industry.
[0114] Various embodiments of the present application are provided above, and the beneficial effects of each embodiment are summarized here:
[0115] 1. Probabilistic structure with state space form. Under the assumption that random noise follows Gaussian distribution, the probabilistic structure in state space form comprehensively considers the dynamic characteristics of variables, the influence of random noise and outliers, significantly improves the interpretability and practicality of the model, and thus improves the accuracy and reliability of anomaly detection results in the tin chemical process.
[0116] 2. Abnormal data elimination based on Hausdorff distance. The maximum difference between data sets is measured by Hausdorff distance, and virtual triggering points and abnormal values in tin chemical process data are accurately eliminated, solving the mutation problem caused by discrete manipulated variables and providing a high-quality data foundation for subsequent dynamic feature extraction and anomaly detection.
[0117] 3. Dynamic feature extraction of PPFA algorithm. The dynamic latent variable characteristics of the tin chemical process are extracted through the PPFA method, the potential dynamic characteristics in the data are captured, parameter estimation is optimized, the model's ability to handle complex industrial process data is enhanced, and the reliability and accuracy of the model are improved.
[0118] 4. Sliding window operation and local sample training. The sliding window operation is used to divide the data into multiple subsequences, and only local samples are used for training to ensure the completeness of the data and the efficiency of training, which is suitable for real-time anomaly detection and early warning.
[0119] You can refer to the following Figure 2 , Figure 2 This is a structural block diagram of a tin chemical process anomaly detection device provided by an embodiment of the present application. The device includes:
[0120] An acquisition module 201 is used to acquire production data collected by sensors in a tin chemical process;
[0121] A preprocessing module 202 is used to preprocess the production data to obtain a target data set corresponding to the production data;
[0122] A construction module 203 is used to construct a state space model based on the target data set, wherein the state space model has a probability structure in a state space form and is used to reflect the mapping relationship between the data of the target data set and the dynamic latent variable characteristics;
[0123] A first calculation module 204 is used to extract dynamic latent variable features corresponding to the training data set through the state space model, and calculate statistical limits according to the dynamic latent variable features corresponding to the training data set;
[0124] A second calculation module 205 is used to extract dynamic latent variable features corresponding to the test data set through the state space model, and calculate statistics according to the dynamic latent variable features corresponding to the test data set;
[0125] The comparison module 206 is used to compare the statistical value with the statistical limit value, and determine whether an abnormality occurs in the tin chemical process according to the comparison result.
[0126] Optionally, the construction module is also used to construct a state space initial model that maps the data of the target data set to dynamic latent variable features through a probabilistic predictable feature analysis (PPFA) algorithm; and estimate and optimize the parameters of the state space initial model to obtain the state space model.
[0127] Optionally, the construction module is also used to estimate and optimize the parameters of the state space initial model by using a maximum likelihood method and / or an expectation maximization algorithm.
[0128] Optionally, the preprocessing module is also used to obtain an original data set corresponding to the production data; divide the original data set into multiple sub-data sets; calculate the Hausdorff distance between the multiple sub-data sets, wherein the Hausdorff distance reflects the similarity between the sub-data sets; eliminate abnormal data in the sub-data sets according to the Hausdorff distance; and integrate the multiple sub-data sets from which the abnormal data has been eliminated to obtain a target data set corresponding to the production data.
[0129] Optionally, the acquisition module is also used to divide the production data into initial stage data, intense stage data and stable stage data according to time information corresponding to the production data; perform noise filtering on the initial stage data to obtain a first data set; perform change feature extraction on the intense stage data to obtain a second data set; use the stable stage data as a third data set; and merge the first data set, the second data set and the third data set to obtain the original data set.
[0130] Optionally, the preprocessing module is further used to perform a sliding window operation on the original data set with a preset window length and a preset sliding step size to obtain a plurality of sub-data sets.
[0131] Optionally, the preprocessing module is further used to calculate the Hausdorff distance between the probability density functions of two adjacent sub-data sets in the window.
[0132] Optionally, the production data includes continuous process data and discrete operation data; the continuous process data includes at least one of the following: temperature, pressure, liquid level; the discrete operation data includes at least one of the following: heating switch, cooling water switch, stirring motor, air intake valve.
[0133] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.
[0134] The embodiment of the present application provides a device for detecting abnormalities in a tin chemical process. Since a state space model is constructed using production data, the state space model has a probability structure in the form of a state space, which is used to reflect the mapping relationship between data and dynamic latent variable characteristics. Therefore, under the assumption that random noise obeys a Gaussian distribution, the solution provided by the embodiment of the present application is superior to deterministic methods (such as dynamic internal principal component analysis (DiPCA) and predictable feature analysis (PFA)). Through the probability structure in the form of a state space, the dynamic characteristics of the variables, the influence of random noise and outliers are fully considered, which significantly improves the interpretability and practicality of the model, thereby improving the accuracy and reliability of the abnormal detection results of the tin chemical process. When used in tin chemical processes, this innovative solution will improve the level of intelligent control of tin chemical production processes and promote the digital transformation of the industry.
[0135] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiment scheme of the present application. Those of ordinary skill in the art can understand and implement it without paying creative work.
[0136] An embodiment of the present application also provides a tin chemical process anomaly detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the tin chemical process anomaly detection method provided in any embodiment of the present application is implemented.
[0137] The above describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0138] As long as there is no conflict or contradiction in the technical features provided in the above embodiments, those skilled in the art can combine the various technical features according to actual conditions to form various different embodiments. However, due to the limited length of this application document, various different embodiments are not described in detail, but it can be understood that various different embodiments also fall within the scope disclosed in the embodiments of this application.
[0139] Those skilled in the art will readily come to other embodiments of the present application after considering the specification and practicing the inventions applied for herein. The present application embodiments are intended to cover any variations, uses or adaptations of the present application embodiments, which follow the general principles of the present application embodiments and include common knowledge or customary technical means in the art that are not applied for by the present application embodiments. The specification and embodiments are to be regarded as exemplary only, and the true scope and spirit of the present application embodiments are indicated by the following claims.
[0140] It should be understood that the embodiments of the present application are not limited to the precise structures described above and shown in the drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the embodiments of the present application is limited only by the appended claims.
[0141] The above description is only a preferred embodiment of the embodiments of the present application and is not intended to limit the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the embodiments of the present application.
Claims
1. A method for detecting abnormality in a tin chemical process, characterized in that: include: Obtain production data collected by sensors in the tin chemical process; Preprocessing the production data to obtain a target data set corresponding to the production data; Building a state space model based on the target data set, wherein the state space model has a probability structure in the form of a state space, and is used to reflect the mapping relationship between the data of the target data set and the dynamic latent variable characteristics; Extracting dynamic latent variable features corresponding to the training data set through the state space model, and calculating statistical limits according to the dynamic latent variable features corresponding to the training data set; Extracting dynamic latent variable features corresponding to the test data set through the state space model, and calculating statistics according to the dynamic latent variable features corresponding to the test data set; The statistical value is compared with the statistical limit value, and whether an abnormality occurs in the tin chemical process is determined based on the comparison result.
2. The method for detecting abnormality in a tin chemical process according to claim 1, characterized in that: The constructing a state space model based on the target data set includes: Constructing an initial state space model mapping the data of the target data set to dynamic latent variable features through a probabilistic predictable feature analysis (PPFA) algorithm; The parameters of the initial state space model are estimated and optimized to obtain the state space model.
3. The method for detecting abnormality in a tin chemical process according to claim 2, characterized in that: The estimating and optimizing the parameters of the initial state space model includes: The parameters of the initial state space model are estimated and optimized by using the maximum likelihood method and / or the expectation maximization algorithm.
4. The method for detecting abnormality in a tin chemical process according to claim 1, characterized in that: The preprocessing of the production data to obtain a target data set corresponding to the production data includes: Obtaining an original data set corresponding to the production data; Dividing the original data set into a plurality of sub-data sets; Calculating the Hausdorff distance between the plurality of sub-datasets, wherein the Hausdorff distance reflects the similarity between the sub-datasets; Eliminating abnormal data in the sub-dataset according to the Hausdorff distance; The plurality of sub-data sets from which abnormal data have been eliminated are integrated to obtain a target data set corresponding to the production data.
5. The method for detecting abnormality in a tin chemical process according to claim 4, characterized in that: The obtaining of the original data set corresponding to the production data includes: According to the time information corresponding to the production data, the production data is divided into initial stage data, severe stage data and stable stage data; Performing noise filtering on the data in the initial stage to obtain a first data set; Extracting change features from the severe stage data to obtain a second data set; Using the stable phase data as the third data set; The first data set, the second data set and the third data set are combined to obtain the original data set.
6. The method for detecting abnormality in a tin chemical process according to claim 4, characterized in that: The step of dividing the original data set into a plurality of sub-data sets comprises: A sliding window operation is performed on the original data set with a preset window length and a preset sliding step length to obtain a plurality of sub-data sets.
7. The method for detecting abnormality in a tin chemical process according to claim 6, characterized in that: The calculating of the Hausdorff distances between the plurality of sub-datasets comprises: The Hausdorff distance between the probability density functions of the two adjacent sub-datasets in the window is calculated.
8. The method for detecting abnormality in a tin chemical process according to claim 1, characterized in that: The production data includes continuous process data and discrete operation data; The continuous process data includes at least one of the following: temperature, pressure, liquid level; The discrete operation data includes at least one of the following: a heating switch, a cooling water switch, a stirring motor, and an air intake valve.
9. A tin chemical process abnormality detection device, characterized in that: include: An acquisition module is used to acquire production data collected by sensors in the tin chemical process; A preprocessing module, used to preprocess the production data to obtain a target data set corresponding to the production data; A construction module, used to construct a state space model based on the target data set, wherein the state space model has a probability structure in the form of a state space, and is used to reflect the mapping relationship between the data of the target data set and the dynamic latent variable characteristics; A first calculation module, configured to extract dynamic latent variable features corresponding to the training data set through the state space model, and calculate statistical limits according to the dynamic latent variable features corresponding to the training data set; A second calculation module is used to extract dynamic latent variable features corresponding to the test data set through the state space model, and calculate statistics according to the dynamic latent variable features corresponding to the test data set; The comparison module is used to compare the statistical value with the statistical limit value, and determine whether an abnormality occurs in the tin chemical process according to the comparison result.
10. A tin chemical process anomaly detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for detecting anomalies in a tin chemical process as described in any one of claims 1 to 8 is implemented.
Citation Information
Patent Citations
Fault detection method and fault detection device for excavator engine
CN111811642A
Pipeline leakage detection method and system based on MPA-VMD denoising
CN115062670A
Fault detection method in tin chemical industry semi-batch process based on dynamic latent variable neighborhood preserving embedding algorithm
CN118427785A
