A method and system for detecting abnormal state of homogenization and viscosity increasing reactor device

Abstract

The present invention relates to the technical field of data processing, and in particular to a method and system for detecting an abnormal state of a homogenizing and viscosity-increasing reactor device. The method can realize information system integration services such as artificial intelligence systems in the production field, artificial intelligence optimization operating systems, artificial intelligence middleware, function libraries and other functions, and can also be used for the development of application software such as computer vision software. The method uses relevant electronic equipment to first obtain a mode association category sequence and an association dictionary sequence corresponding to a reactor device under each normal working mode, obtains sample data of the reactor device in actual work in real time, and compares the data under each normal working mode to obtain the actual working mode, and judges whether the reactor device has an abnormal state according to whether the corresponding actual working mode can be obtained; the correlation between different categories of data under each working mode is combined to improve the accuracy of data analysis and increase the credibility of abnormal state detection of the reactor device.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for detecting an abnormal state of a homogenizing and viscosity-increasing reactor device. Background Art

[0002] In the operation of modern chemical industry, various data information of equipment in the production process is usually collected, and the equipment is monitored based on the relationship between a large amount of data in the database and real-time sample data. The chemical production process usually includes multiple production modes, and there is a certain correlation between the data of each production mode.

[0003] Existing analyses of multimodal chemical processes often only consider the uniqueness of the data corresponding to each production mode, or consider the correlation between different production modes on the basis of considering the uniqueness of each production mode; however, for the same production mode, the connection between the data under the production mode is often closer, and abnormal conditions may occur in the production process due to the mutual influence between the data. Ignoring the correlation between the data under the same production mode for analysis may result in deviations and low accuracy in detection and identification of the device. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a method and system for detecting abnormal state of a homogenizing and viscosity increasing reactor device. The technical scheme adopted is as follows:

[0005] In a first aspect, an embodiment of the present invention provides a method for detecting abnormal state of a homogenizing and viscosity increasing reactor device, the method comprising the following steps:

[0006] Collecting a plurality of sample data corresponding to the reactor device in each normal working mode, wherein the sample data includes a plurality of category data of the reactor device, and each category data is a data sequence of the reactor device in the corresponding normal working mode;

[0007] Obtaining the degree of correlation between any two of the category data in each of the normal working modes, constructing a mode association category sequence corresponding to the normal working mode according to the degree of correlation, and obtaining a corresponding association dictionary sequence according to the mode association category sequence;

[0008] Collecting a plurality of actual sample data of the reactor device during actual operation, and obtaining an actual pattern association category sequence and an actual association dictionary sequence according to the plurality of actual sample data; obtaining a similarity between the actual pattern association category sequence and each of the pattern association category sequences, wherein the pattern association category sequence when the similarity is greater than a preset threshold is a candidate pattern association category sequence, and each of the candidate pattern association category sequences corresponds to a candidate association dictionary sequence;

[0009] Acquire the similarity between the actual associated dictionary sequence and each of the candidate associated dictionary sequences, the candidate associated dictionary sequence corresponding to the maximum similarity is the optimal sequence, and the normal working mode corresponding to the optimal sequence is the actual working mode during actual working;

[0010] The actual working mode of the reactor device in multiple consecutive time periods is obtained, and the time period without the actual working mode is the suspected abnormal time. The suspected abnormal time is compared with the two adjacent time periods before and after it to determine whether an abnormal state occurs during the suspected abnormal time.

[0011] Preferably, the step of obtaining the correlation between any two of the category data in each of the normal working modes includes:

[0012] The difference between any two of the category data in each of the sample data is obtained, and the mean of the difference between the two category data in all the sample data is calculated, where the mean of the difference is the correlation degree between the two category data.

[0013] Preferably, the step of constructing a mode association category sequence corresponding to the normal working mode according to the correlation degree includes:

[0014] The correlation degree between any two of the category data in each of the normal working modes is obtained, and all the correlation degrees are arranged in descending order to obtain a corresponding mode-associated category sequence.

[0015] Preferably, the step of obtaining a corresponding association dictionary sequence according to the pattern association category sequence includes:

[0016] A data matrix is ​​constructed according to each element in the pattern association category sequence, and each of the data matrices is decomposed to obtain a dictionary matrix, and all the dictionary matrices constitute a corresponding association dictionary sequence.

[0017] Preferably, the step of obtaining the similarity between the actual pattern association category sequence and each of the pattern association category sequences comprises:

[0018] The maximum distance value corresponding to each element in the actual pattern association category sequence is obtained, and the weight of each element in the actual pattern association category sequence is obtained based on the difference between the element value at each position in the actual pattern association category sequence and the element value at the same position in any of the pattern association category sequences; and the similarity between the actual pattern association category sequence and the pattern association category sequence is obtained according to the maximum distance value and the weight.

[0019] Preferably, the step of obtaining the similarity between the actual associated dictionary sequence and each of the candidate associated dictionary sequences includes:

[0020] Obtain the entropy value of each element in the actual associated dictionary sequence to obtain a corresponding actual entropy sequence; obtain the entropy value of each element in each of the candidate associated dictionary sequences to obtain a corresponding candidate entropy sequence; calculate the cosine value between the actual entropy sequence and each of the candidate entropy sequences as the similarity between the actual associated dictionary sequence and each of the candidate associated dictionary sequences.

[0021] Preferably, the step of comparing the suspected abnormal time with the two adjacent time periods before and after it to determine whether an abnormal state has occurred at the suspected abnormal time includes:

[0022] Obtaining actual sample data for each time period, and obtaining actual working modes corresponding to the two time periods before and after the suspected abnormal time; calculating the gap between the actual sample data of each time period and each normal working mode, and obtaining the exit mode and response mode of the suspected abnormal time based on the gap corresponding to each time period;

[0023] A response rate corresponding to the suspected abnormal time is obtained according to the transfer-out mode and the response mode. When the response rate is less than a preset threshold, the reactor device has an abnormal state at the suspected abnormal time.

[0024] In a second aspect, a system for detecting abnormal conditions of a homogenizing and viscosity increasing reactor device is provided, the system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps described in any one of the above methods when executing the computer program.

[0025] The present invention has the following beneficial effects: the embodiments of the present invention can realize information system integration services such as artificial intelligence systems in the production field, artificial intelligence optimization operating systems, artificial intelligence middleware, function libraries and other functions, and can also be used for the development of application software such as computer vision software; by analyzing the degree of correlation between different categories of data in each different working mode, the accuracy of data analysis is increased; the correlation between each category of data in different working modes is combined to more accurately locate the abnormal state of the reactor device, and the possibility of working mode conversion is eliminated during the analysis process, thereby reducing errors and improving the accuracy of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 A flow chart of a method for detecting abnormal state of a homogenizing and viscosity increasing reactor device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the abnormal state detection method and system of a homogenization viscosity-increasing reactor device proposed by the present invention, its specific implementation method, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0029] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0030] The present application is suitable for detecting abnormal conditions during the operation of a reactor device. By analyzing the correlation between different categories of data when the reactor device is operating, the actual operating mode during real-time operation is acquired, thereby identifying the time period in which the abnormal condition may occur, and judging whether the abnormal condition has occurred based on the difference between the sample data of the time period and the adjacent time period, thereby effectively improving the final detection result.

[0031] The specific scheme of the abnormal state detection method and system of a homogenizing and viscosity increasing reactor device provided by the present invention is described in detail below with reference to the accompanying drawings.

[0032] See also Figure 1 , which shows a flow chart of a method for detecting abnormal state of a homogenizing and viscosity increasing reactor device provided by an embodiment of the present invention, the method comprising the following steps:

[0033] Step S100 , collecting a plurality of sample data corresponding to each normal working mode of the reactor device, wherein the sample data includes a plurality of category data of the reactor device, and each category data is a data sequence of the reactor device in the corresponding normal working mode.

[0034] Since chemical plants may have problems with energy scheduling or order fluctuations, for example, due to the high energy consumption of chemical companies, they usually adjust the production output during the day and night according to the difference in industrial electricity prices in the morning and evening. Therefore, chemical objects do not always operate in a fixed production mode; whether in normal or abnormal operating conditions, the equipment in chemical plants may have multiple production modes, which is called a multimodal chemical process.

[0035] In the embodiment of the present invention, the abnormal state of the homogenization and viscosity increasing reactor device in the multimodal chemical process is identified. During normal use, the reactor device mainly includes three types of data: temperature data, flow data and pressure data. In the embodiment of the present invention, the three types of data are subdivided into 7 categories of data, namely, feed flow data, reactor pressure data, reactor liquid level data, reactor temperature data, reactor feed valve opening data, reactor condensate flow data and condensate cooling water flow data.

[0036] It should be noted that, during the actual use of the reactor device, the corresponding category data is different in different production modes, and different production modes correspond to different working modes.

[0037] Furthermore, data of the reactor device in a chemical process under multiple normal operating modes are collected, and the acquired data are recorded as sample data under the normal operating mode. The sample data include data sequences of all categories of the reactor device under the normal operating mode, and each data sequence is a sequence of sampled data from the beginning to the end of the normal operating mode; that is, there are several categories of data of the reactor device under the normal operating mode, and the sample data include several categories of data sequences.

[0038] In this way, the sample data corresponding to each normal working mode of the reactor device is obtained.

[0039] Step S200, obtaining the correlation between any two category data in each normal working mode, constructing a mode association category sequence corresponding to the normal working mode according to the correlation, and obtaining a corresponding association dictionary sequence according to the mode association category sequence.

[0040] Under different normal working modes, different categories of data in the reactor device often influence each other. For example, an increase in flow data may lead to an increase in pressure data. Therefore, there are different degrees of correlation between different categories of data. The correlation between different categories of data is analyzed.

[0041] Since each sample data includes multiple categories of data, and each category of data is a data sequence, different categories of data in the same sample data are collected in the same time period. In order to improve the accuracy of the analysis, multiple sample data are collected in each normal working mode; the correlation between different categories of data is analyzed based on the multiple sample data in the normal working mode. The difference between any two categories of data in each sample data is obtained, and the mean of the difference between the two categories of data in all sample data is calculated. The mean of the difference is the correlation between the two categories of data.

[0042] Specifically, the difference between any two categories of data in each sample data is obtained. In the embodiment of the present invention, the DTW algorithm is used to analyze two different categories of data, and the DTW distance between any two categories of data in the sample data is obtained, and all the obtained DTW distances are normalized. Similarly, the DTW distance between any two categories of data in each sample data under the same normal working mode is obtained; the mean DTW distance of all sample data of any two categories of data under the same normal working mode is calculated, and the mean DTW distance is used as the correlation between the two categories of data under the normal working mode; based on the same method, the correlation between any two categories of data under each normal working mode is obtained.

[0043] Further, the correlation between any two categories of data in each normal working mode is obtained, and all the correlations are arranged in descending order to obtain a corresponding mode-associated category sequence. That is, the correlation between all the categories of data in each normal working mode is sorted, and in the embodiment of the present invention, they are arranged in descending order according to the correlation, thereby forming a mode-associated category sequence corresponding to the normal working mode, and each element in the mode-associated category sequence is the correlation between any two categories of data in the normal working mode.

[0044] In order to more accurately analyze the relationship characteristics between each category data, the associated dictionary sequence corresponding to each pattern-associated category sequence is obtained; each element in the pattern-associated category sequence is the degree of correlation between two category data, that is, each element is a category data combination, and each category data combination corresponds to a dictionary to construct an associated dictionary sequence. A data matrix is ​​constructed according to each element in the pattern-associated category sequence, and each data matrix is ​​decomposed into a dictionary matrix. All dictionary matrices constitute the corresponding associated dictionary sequence.

[0045] Specifically, the method for obtaining the dictionary corresponding to each category data combination is as follows:

[0046] (1) Convert different categories of data into data matrices.

[0047] As a preferred example, if any sample data includes three category data, namely category data a, category data b and category data c, and each category data is a data sequence; then there are three category data combinations in the sample data, namely ab, ac and bc.

[0048] Since multiple sample data are collected in each normal working mode, assuming that three sample data are collected in each normal working mode, the three sample data respectively include category data a, category data b and category data c; in the embodiment of the present invention, a data interleaving method is used to construct a data matrix of different category data combinations. For category data combination ab, since the three sample data include category data combination ab, the data matrix corresponding to the category data combination is:

[0049]

[0050] Among them, n 1 Represents the data matrix corresponding to the first category data combination; a1 represents the category data a in the first sample data; b1 represents the category data b in the first sample data; a2 represents the category data a in the second sample data; b2 represents the category data b in the second sample data; a3 represents the category data a in the third sample data; b3 represents the category data b in the third sample data.

[0051] And so on, obtain the data matrix corresponding to each category data combination.

[0052] (2) Based on the K-SVD algorithm, the data matrix corresponding to each category data combination is decomposed.

[0053] The K-SVD algorithm is an algorithm for dictionary learning in sparse representation. Its name comes from the fact that the algorithm has to go through K iterations, and each iteration uses SVD decomposition. The algorithm can decompose a digital matrix into a dictionary matrix and a sparse matrix. The dictionary matrix is ​​used to represent the most essential features of the digital matrix, and the sparse matrix represents the combination of different features in the dictionary matrix.

[0054] Therefore, the data matrix corresponding to each category data combination is decomposed into a corresponding dictionary matrix and a sparse matrix through the K-SVD algorithm. The specific decomposition method is a well-known technology and will not be repeated here.

[0055] (3) Construct an associated dictionary sequence.

[0056] The data matrix corresponding to each category data combination is decomposed into a corresponding dictionary matrix and a sparse matrix through the above-mentioned K-SVD algorithm; thus, a dictionary matrix corresponding to each category data combination in each normal working mode is obtained, and then an associated dictionary sequence is constructed according to the dictionary matrix corresponding to each category data combination. The associated dictionary sequence corresponds one-to-one to the elements in the pattern-associated category sequence, that is, the first dictionary matrix in the associated dictionary sequence is obtained by decomposing the data matrix corresponding to the first category data combination in the pattern-associated category sequence.

[0057] In this way, the associated dictionary sequence and mode-associated category sequence corresponding to each normal working mode are obtained.

[0058] Step S300, collecting multiple actual sample data of the reactor device during actual operation, obtaining an actual pattern association category sequence and an actual association dictionary sequence based on the multiple actual sample data; obtaining the similarity between the actual pattern association category sequence and each pattern association category sequence, and the pattern association category sequence when the similarity is greater than a preset threshold is a candidate pattern association category sequence, and each candidate pattern association category sequence corresponds to a candidate association dictionary sequence.

[0059] Specifically, the mode association category sequence and the association dictionary sequence corresponding to each normal working mode of the reactor device are obtained in step S200; since the actual sample data collected during the actual working process of the reactor device cannot determine its corresponding working mode, the corresponding actual working mode is obtained by analyzing the similarity between the actual sample data actually collected and the sample data under multiple normal working modes, that is, the analysis is performed based on the similarity between the actual mode association category sequence and the actual association dictionary sequence corresponding to the actual sample data and the mode association category sequence and the association dictionary sequence under multiple normal working modes. The maximum distance value corresponding to each element in the actual mode association category sequence is obtained, and the weight of each element in the actual mode association category sequence is obtained based on the difference between the element value at each position in the actual mode association category sequence and the element value at the same position in any mode association category sequence, and the similarity between the actual mode association category sequence and the mode association category sequence is obtained based on the maximum distance value and the weight.

[0060] First, obtain the maximum distance value of each element in the actual pattern association category sequence; the maximum distance value is obtained by the position of each element. As an example, assuming that the actual pattern association category sequence A = {a, b, c, d}, the position of the first element a in the actual pattern association category sequence is 1; and the maximum position value that the element a can exist in the actual pattern association category sequence A is 4, then the maximum distance value corresponding to element a is 4-1 = 3. In this way, the maximum distance value corresponding to element b is 2, the maximum distance value corresponding to element c is 2, and the maximum distance value corresponding to element d is 3.

[0061] And so on, the maximum distance value corresponding to each element in the actual pattern association category sequence is obtained.

[0062] Then, the actual distance value of each element in the actual mode association category is obtained, and the actual distance value is obtained by the element position of the actual mode association category sequence and the corresponding element position in the mode association category sequence under each normal working mode. As an example, assuming that the actual mode association category sequence is A = {a, b, c, d}, where each element represents a category data combination; the mode association category sequence under any normal working mode is B = {c, a, b, d}, the position of element a in the actual mode association category sequence is 1; the position of element a in the mode association category sequence under any normal working mode is 2, then the actual distance value of element a in the actual mode association category sequence is 2-1 = 1; the actual distance value of element b in the actual mode association category sequence is 1; the actual distance value of element c in the actual mode association category sequence is 2; the actual distance value of element d in the actual mode association category sequence is 0.

[0063] By analogy, the actual distance value corresponding to each element in the actual mode association category sequence is obtained. According to the actual distance value corresponding to each element in the actual mode association category sequence and the maximum distance value, the distance value of each element in the actual mode association category sequence and the mode association category sequence in the normal working mode is obtained; in the embodiment of the present invention, the distance value of each element is set to the ratio of the actual distance value corresponding to the element to the maximum distance value.

[0064] Finally, a weight is assigned to each element in the actual pattern association category sequence, and the weight of each element is obtained by the difference between the value of the element in the actual pattern association category and the value of the element at the same position in the normal pattern association category sequence. As a preferred example, assume that the actual pattern association category sequence A = {a, b, c, d}, and the pattern association category sequence in any normal working mode is B = {c, a, b, d}; for element a in the actual pattern association category sequence, element a is in the first position in the actual pattern association category sequence, so the weight of element a is the absolute value of the difference between the first position element c in the normal pattern association category sequence B and element a; the value of each element is the degree of correlation between the corresponding category data combination. Then the weight of element b in the actual pattern association category sequence A is |ab|; the weight of element c in the actual pattern association category sequence A is |bc|; the weight of element d in the actual pattern association category sequence A is |dd|.

[0065] Similarly, the weight corresponding to each element in the actual pattern association category sequence is obtained, and the similarity between the actual pattern association category sequence and the normal pattern association category sequence is obtained by combining the distance value of each element in the actual pattern association category sequence. The specific similarity calculation method is: calculate the weighted sum of the distance value and the weight of each element in the actual pattern association category sequence, normalize the result of the weighted sum, and the similarity is the inverse of the weighted sum result.

[0066] By analogy, the similarity between the actual mode association category sequence and the mode association category sequence under each normal working mode is obtained. Since there is a certain correlation between different working modes, the actual mode association category sequence may be consistent with the similarity between the mode association category sequences under multiple normal working modes, and it is impossible to determine which normal working mode the actual mode association category sequence belongs to. In the embodiment of the present invention, the normal working modes whose similarity with the actual mode association category sequence is greater than a preset threshold are screened and retained, and the association dictionary sequences corresponding to the multiple normal working modes retained are further analyzed with the actual association dictionary sequences of the actual sample data.

[0067] Preferably, in an embodiment of the present invention, a preset threshold is set at 0.9, and pattern association category sequences having a similarity with the actual pattern association category sequence greater than 0.9 are retained, that is, multiple normal working modes similar to the actual work are retained, and the retained normal working modes are recorded as candidate working modes, and each candidate working mode corresponds to a candidate association dictionary sequence.

[0068] Step S400, obtaining the similarity between the actual associated dictionary sequence and each candidate associated dictionary sequence, the candidate associated dictionary sequence corresponding to the maximum similarity is the optimal sequence, and the normal working mode corresponding to the optimal sequence is the actual working mode during actual working.

[0069] Furthermore, the similarity between the actual associated dictionary sequence of the actual work and multiple candidate associated dictionary sequences is analyzed, and the entropy value of each element in the actual associated dictionary sequence is obtained to obtain the corresponding actual entropy sequence; the entropy value of each element in each candidate associated dictionary sequence is obtained to obtain the corresponding candidate entropy sequence; and the cosine value between the actual entropy sequence and each candidate entropy sequence is calculated as the similarity between the actual associated dictionary sequence and each candidate associated dictionary sequence.

[0070] Since each element in the associated dictionary sequence is a dictionary matrix, the entropy value of each element in the actual associated dictionary sequence of the actual work is obtained, and the actual entropy sequence corresponding to the actual associated dictionary sequence is obtained according to the entropy value of each element. Accordingly, the entropy value of each element in the candidate associated dictionary sequence corresponding to multiple candidate working modes is obtained, so as to obtain the candidate entropy sequence corresponding to each candidate associated dictionary sequence.

[0071] In the embodiment of the present invention, the cosine value between the actual entropy sequence and each candidate entropy sequence is obtained as the corresponding similarity. Thus, the similarity between the actual entropy sequence of the actual work and the candidate entropy sequence of each candidate working mode is obtained. The candidate entropy sequence with the greatest similarity to the actual entropy sequence is selected as the optimal entropy sequence, and the candidate working mode corresponding to the optimal entropy sequence is the actual working mode of the actual work.

[0072] Based on the above method for analyzing the actual working mode of the actual work, the corresponding actual working mode at this time can be obtained according to the actual sample data obtained in real time.

[0073] Step S500, obtaining the actual working mode of the reactor device in multiple consecutive time periods, and the time period without the actual working mode is the suspected abnormal time. The suspected abnormal time is compared with the two adjacent time periods before and after it to determine whether an abnormal state occurs during the suspected abnormal time.

[0074] Specifically, the actual working mode corresponding to the reactor device when working in real time is obtained by the method in step S300 and step S400; in the existing process of detecting the abnormal state of the reactor device, the state when switching between different working modes is often determined to be an abnormal state. This method often has a large error in detecting the abnormal state. In the embodiment of the present invention, the abnormal state of the reactor device is detected by the actual working mode corresponding to the reactor in multiple time periods.

[0075] As a preferred example, in the embodiment of the present invention, data collection is performed when the reactor device starts to run, and the time interval for data collection is set to half an hour, that is, sample data of the reactor device is collected every half an hour. Assuming that the reactor device has been running for a total of two hours up to this moment, the actual sample data of the four time periods corresponding to the reactor device are obtained. The actual sample data of each time period is matched with the sample data under multiple normal working modes to obtain the actual working mode corresponding to the actual sample data of each time period. When there is a time period in the four time periods that does not match the actual working mode, the time period may be an abnormal state of the reactor device, or the reactor device may have undergone a change in working mode during the time period, so the actual sample data of the time period is further analyzed.

[0076] Acquire the actual sample data of each time period, and acquire the actual working modes corresponding to the two time periods before and after the suspected abnormal time; calculate the gap between the actual sample data of each time period and each normal working mode, and obtain the roll-out mode and response mode of the suspected abnormal time based on the gap corresponding to each time period; obtain the response rate corresponding to the suspected abnormal time according to the roll-out mode and the response mode, and when the response rate is less than the preset threshold value, the reactor device has an abnormal state at the suspected abnormal time.

[0077] Specifically, the time period where the abnormal state may occur is marked as the i-th time period, and whether the abnormal state occurs is determined by analyzing the sample data corresponding to the i-1-th time period, the i-th time period, and the i+1-th time period.

[0078] First, obtain the curve corresponding to each time period, which is obtained from the actual sample data corresponding to the time period; from the method of obtaining the pattern association category sequence and the associated dictionary sequence of the sample data in step S200, it can be seen that every two category data combinations correspond to a data matrix, and each data matrix can be decomposed into a dictionary matrix and a sparse matrix; calculate the entropy value of each dictionary matrix and the entropy value of each sparse matrix, and take the entropy value of the dictionary matrix and the entropy value of the sparse matrix as a coordinate point (E 1 ,E 2 );E 1 Represents the entropy value of the dictionary matrix; E 2Represents the entropy value of the sparse matrix. Similarly, the coordinate points corresponding to each combination of two category data in each actual sample data are obtained, and the lines between all the coordinate points constitute the curve corresponding to the actual sample data.

[0079] Then, based on the same method as the above method for obtaining the corresponding curve of the actual sample data of each time period, the curve of the sample data corresponding to each normal working mode is obtained, and the DTW distance between the corresponding curve of the actual sample data of each time period and the corresponding curve of each normal working mode is calculated, and each DTW distance is used as the gap between the actual mode of the corresponding time period and each normal working mode. Since there is no definite working mode in the i-th time period, the transfer-out rate and transfer-in rate of the i-th time period between multiple normal working modes in the i-1th time period are calculated.

[0080] The calculation method of the transfer-out rate is: obtain the transfer-out mode in the i-th time period, and the transfer-out mode is that the gap corresponding to the i-th time period under a certain normal working mode is larger than the gap corresponding to the i-1-th time period; then the transfer-out rate corresponding to the transfer-out mode is the difference between the gap corresponding to the i-th time period and the gap corresponding to the i-1-th time period.

[0081] The calculation method of the transfer-in rate is: obtain the response mode in the i-th time period, and the response mode is that the gap corresponding to the i-th time period under a certain normal working mode is smaller than the gap corresponding to the i-1-th time period; then the transfer-in rate corresponding to the response mode is the difference between the gap corresponding to the i-1-th time period and the gap corresponding to the i-th time period.

[0082] Finally, the response rate corresponding to the i-th time period is used for final analysis. The response rate is the ratio of the transfer-out rate to the transfer-in rate in the i-th time period, and it is determined whether the response rate is greater than the preset threshold. At the same time, the actual sample data of the i+1-th time period is analyzed to determine whether there is a corresponding normal working mode in the time period. When the i+1-th time period has an actual working mode corresponding to the normal working mode and the response rate of the i-th time period is greater than the preset threshold, it is considered that no abnormal state has occurred in the i-th time period; otherwise, it is considered that an abnormal state has occurred in the time period.

[0083] Preferably, in the embodiment of the present invention, the preset threshold of the response rate is set to 0.8.

[0084] As an example, please refer to Table 1, which shows a reference table of gap data between multiple time periods and different normal working modes.

[0085] Table 1

[0086] Normal operation mode A Normal operation mode B Normal operation mode C i-1th time period 0.1 0.8 0.9 The i-th time period 0.6 0.8 0.4 i+1th time period 0.9 0.8 0.1

[0087] Each data represents the difference between the curve of the actual sample data of the corresponding time period and the curves of different normal working modes; for the i-1th time period, the difference between it and the normal working mode A is the smallest, so the most likely actual mode of this time period is the normal working mode A; for the i-th time period, the difference between it and each normal working mode is large, and there is a transfer-out mode and a response mode in this time period; the transfer-out rate corresponding to the transfer-out mode in this time period is obtained as: 0.6-0.1=0.5; accordingly, the transfer-in rate of the response mode in this time period is calculated as: 0.9-0.4=0.6; based on the transfer-out rate and transfer-in rate obtained in this time period, the response rate corresponding to the i-th time period is obtained as: The response rate in this time period is greater than the preset threshold value of 0.8. Further judging that the gap between the i+1th time period and the normal working mode C is the smallest, so the most likely actual mode of this time period is the normal working mode C. It can be determined that there is a transition of working modes between the i-1th time period and the i+1th time period. Although the i-th time period does not match the corresponding normal working mode, this time period is not in an abnormal state, but is only transitioning between working modes.

[0088] In summary, in the embodiment of the present invention, the correlation between different categories of data in each normal working mode is analyzed to obtain the mode association category sequence and the association dictionary sequence corresponding to each normal working mode; the actual sample data during real-time working is analyzed to obtain the corresponding normal working mode. When the reactor device does not have the corresponding actual working mode in a certain time period of the working process, the sample data in the time period is analyzed again to determine whether the time period is a transition between working modes or an abnormal state occurs. Considering the correlation between different categories of data makes the analysis more accurate, and the final result of the abnormal state judgment is more reliable.

[0089] Based on the same inventive concept as the above method embodiment, the present invention also provides a system for detecting abnormal state of a homogenizing and viscosity-increasing reactor device, the system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above method embodiment for detecting abnormal state of a homogenizing and viscosity-increasing reactor device are implemented, for example Figure 1 The abnormal state detection method of the homogenization and viscosity increasing reactor device has been described in detail in the above embodiment and will not be repeated here.

[0090] It should be noted that the sequence of the embodiments of the present invention described above is for description only and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of this specification. 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 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.

[0091] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting abnormal state of a homogenizing and viscosity increasing reactor device, characterized in that: The method comprises the following steps: Collecting a plurality of sample data corresponding to the reactor device in each normal working mode, wherein the sample data includes a plurality of category data of the reactor device, and each category data is a data sequence of the reactor device in the corresponding normal working mode; Obtaining the degree of correlation between any two of the category data in each of the normal working modes, constructing a mode association category sequence corresponding to the normal working mode according to the degree of correlation, and obtaining a corresponding association dictionary sequence according to the mode association category sequence; Collecting a plurality of actual sample data of the reactor device during actual operation, and obtaining an actual pattern association category sequence and an actual association dictionary sequence according to the plurality of actual sample data; obtaining a similarity between the actual pattern association category sequence and each of the pattern association category sequences, wherein the pattern association category sequence when the similarity is greater than a preset threshold is a candidate pattern association category sequence, and each of the candidate pattern association category sequences corresponds to a candidate association dictionary sequence; Acquire the similarity between the actual associated dictionary sequence and each of the candidate associated dictionary sequences, the candidate associated dictionary sequence corresponding to the maximum similarity is the optimal sequence, and the normal working mode corresponding to the optimal sequence is the actual working mode during actual working; The actual working mode of the reactor device in multiple consecutive time periods is obtained, and the time period without the actual working mode is the suspected abnormal time. The suspected abnormal time is compared with the two adjacent time periods before and after it to determine whether an abnormal state occurs during the suspected abnormal time.

2. The method for detecting abnormal state of a homogenizing and viscosity increasing reactor device according to claim 1, characterized in that: The step of obtaining the correlation between any two of the category data in each of the normal working modes comprises: The difference between any two of the category data in each of the sample data is obtained, and the mean of the difference between the two category data in all the sample data is calculated, where the mean of the difference is the correlation degree between the two category data.

3. The method for detecting abnormal state of a homogenizing and viscosity increasing reactor device according to claim 1, characterized in that: The step of constructing the mode association category sequence corresponding to the normal working mode according to the correlation degree comprises: The correlation degree between any two of the category data in each of the normal working modes is obtained, and all the correlation degrees are arranged in descending order to obtain a corresponding mode-associated category sequence.

4. The method for detecting abnormal state of a homogenizing and viscosity increasing reactor device according to claim 1, characterized in that: The step of obtaining a corresponding association dictionary sequence according to the pattern association category sequence comprises: A data matrix is ​​constructed according to each element in the pattern association category sequence, and each of the data matrices is decomposed to obtain a dictionary matrix, and all the dictionary matrices constitute a corresponding association dictionary sequence.

5. The method for detecting abnormal state of a homogenizing and viscosity increasing reactor device according to claim 1, characterized in that: The step of obtaining the similarity between the actual pattern association category sequence and each of the pattern association category sequences comprises: The maximum distance value corresponding to each element in the actual pattern association category sequence is obtained, and the weight of each element in the actual pattern association category sequence is obtained based on the difference between the element value at each position in the actual pattern association category sequence and the element value at the same position in any of the pattern association category sequences; and the similarity between the actual pattern association category sequence and the pattern association category sequence is obtained according to the maximum distance value and the weight.

6. The method for detecting abnormal state of a homogenizing and viscosity increasing reactor device according to claim 1, characterized in that: The step of obtaining the similarity between the actual associated dictionary sequence and each of the candidate associated dictionary sequences comprises: Obtain the entropy value of each element in the actual associated dictionary sequence to obtain a corresponding actual entropy sequence; obtain the entropy value of each element in each of the candidate associated dictionary sequences to obtain a corresponding candidate entropy sequence; calculate the cosine value between the actual entropy sequence and each of the candidate entropy sequences as the similarity between the actual associated dictionary sequence and each of the candidate associated dictionary sequences.

7. The method for detecting abnormal state of a homogenizing and viscosity increasing reactor device according to claim 1, characterized in that: The step of comparing the suspected abnormal time with the two adjacent time periods before and after it to determine whether an abnormal state has occurred during the suspected abnormal time includes: Obtaining actual sample data for each time period, and obtaining actual working modes corresponding to the two time periods before and after the suspected abnormal time; calculating the gap between the actual sample data of each time period and each normal working mode, and obtaining the exit mode and response mode of the suspected abnormal time based on the gap corresponding to each time period; A response rate corresponding to the suspected abnormal time is obtained according to the transfer-out mode and the response mode. When the response rate is less than a preset threshold, the reactor device has an abnormal state at the suspected abnormal time.

8. A system for detecting abnormal state of a homogenizing and viscosity increasing reactor, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When executing the computer program, the processor implements the steps of the method according to any one of claims 1 to 7.

Images