A method and device for determining distillation tower failure

Abstract

The present invention discloses a method and device for determining a distillation tower fault, which relates to the field of industrial process detection and control, and is applied to a distillation tower, including determining the inverse covariance matrix of the historical operating data of the distillation tower; calculating a preset sample reachable value based on the inverse covariance matrix of the historical operating data; and determining the inverse covariance matrix of the operating data of the current cycle. According to the historical operating data, an inverse covariance matrix for determining the relationship between various influencing variables and a maximum value that can be reached by the historical operating data as a whole, i.e., a preset sample reachable value, can be obtained. When the operating data of the current cycle is obtained, the sample reachable value of the current cycle is compared with the preset sample reachable value to determine whether a fault currently occurs. When a fault occurs, the cause of the fault, i.e., the variable that causes the alarm, is determined based on the inverse covariance matrix of the operating data of the current cycle. An alarm is given in time for the occurrence of a fault, and the cause of the fault can be determined, and the fault can be accurately located.

Description

Technical Field

[0001] The invention relates to the field of industrial process detection and control, and in particular to a method and device for determining a distillation tower fault. Background Art

[0002] In the field of process industry, distillation tower is a common equipment. Taking advantage of the different boiling points of different substances, distillation tower is generally used to separate and purify mixed raw materials. The distillation tower maintains a certain temperature and pressure inside the distillation tower by heating the tower kettle, and then produces the target component at the top of the tower. In order to ensure the purity and stability of the target component, it is very important to maintain the smooth operation of the distillation tower. However, with the increase of the operation time of the production equipment, due to equipment aging, failure, impurities in the materials, etc., the distillation tower may have problems such as tray blockage and tower body damage, which will affect the operation efficiency of the distillation tower and even affect production safety. The failure of the distillation tower is generally caused by multiple variables such as temperature, pressure, flow, etc. Due to the mutual influence between these multiple variables, the relevant technology cannot accurately give the root cause variable that causes the failure of the distillation tower, that is, to find the most important variable from temperature, pressure, flow, etc., which causes the monitoring statistic to exceed the limit. Therefore, it is of great significance to establish a method that can accurately describe the interaction relationship between variables. Summary of the invention

[0003] The purpose of the present invention is to provide a method and device for determining a distillation tower fault, determine the specific cause of the fault, that is, the variable causing the alarm, give an alarm in time for the occurrence of the fault, determine the cause of the fault, and accurately locate the fault.

[0004] In order to solve the above technical problems, the present invention provides a method for determining a distillation tower failure, which is applied to a distillation tower, wherein the upper part of the distillation tower is a tower top, the lower part of the distillation tower is a tower bottom, and raw materials are input from the middle part of the distillation tower and output from the tower top and the tower bottom respectively;

[0005] The method for determining the distillation tower failure comprises:

[0006] Determine an inverse covariance matrix of historical operating data of the distillation tower, wherein the operating data includes a combination of two or more of the pressure at the tower top, the temperature at the tower top, the temperature at the tower bottom, the liquid level height at the tower bottom, and the temperature in the middle of the distillation tower, wherein the numerical values ​​in the inverse covariance matrix represent a causal relationship between any two of the operating data, and the numerical values ​​are positively correlated with the causal relationship;

[0007] Calculating a preset sample reachable value according to an inverse covariance matrix of the historical operation data, wherein the sample reachable value represents a maximum value of the operation data of the distillation tower;

[0008] Determine the inverse covariance matrix of the running data for the current period;

[0009] Determine the sample reachable value of the current period according to the inverse covariance matrix of the current period;

[0010] When the sample reachable value of the current period is greater than the preset sample reachable value, an alarm is issued;

[0011] The variable causing the alarm in the operation data is determined according to the inverse covariance matrix of the current cycle.

[0012] On the other hand, determining the inverse covariance matrix of the historical operating data of the distillation column comprises:

[0013] Determining a mean of the historical operating data;

[0014] According to the historical operation data, the mean of the historical operation data and the first relational expression θ=(X-μ)(X-μ) T determining an inverse covariance matrix of the historical operating data;

[0015] Wherein θ is the inverse covariance matrix of the historical operating data, X is the historical operating data, and μ is the mean of the historical operating data.

[0016] On the other hand, before determining the inverse covariance matrix of the historical operating data of the distillation tower, the method further includes:

[0017] Using a k-means clustering algorithm to classify the historical operating data of the distillation tower into N types of operating points;

[0018] Determine the probability of occurrence of each work point category;

[0019] Determine a weight of each of the historical operating data belonging to each category according to the occurrence probability and the inverse covariance matrix of the historical operating data of each category, wherein the weight represents the possibility that the historical operating data belongs to each operating point category;

[0020] The operating point category corresponding to the historical operating data is updated according to the weight.

[0021] On the other hand, after determining the inverse covariance matrix of the running data of the current period, it also includes:

[0022] Determine the weight of each category of the operation data of the current period according to the inverse covariance matrix of the operation data of the current period;

[0023] Using the category with the largest weight as the category of the operation data of the current cycle;

[0024] When the sample reachable value of the current period is greater than the preset sample reachable value, an alarm is issued, including:

[0025] An alarm is issued when the sample reachable value of the current period is greater than the preset sample reachable value corresponding to the category of the running data of the current period.

[0026] On the other hand, determining the weight of each of the historical operating data belonging to each category according to the occurrence probability and the inverse covariance matrix of the historical operating data of each category includes:

[0027] Determine an upper triangular matrix L of the inverse covariance matrix of the historical operating data;

[0028] According to the upper triangular matrix, the historical operation data and the second relational expression Determine a summation value q, wherein the summation value represents the insignificance of each of the historical operating data under the inverse covariance matrix of the current operating point category, and the closer q is to 0, the more significantly the historical operating data belongs to the current operating point category, and X is the historical operating data;

[0029] Determine a discriminant value c according to the upper triangular matrix and the third relational expression c=log22π+2||log(diag(L))||1, wherein the discriminant value represents the overall insignificance of all the historical operating data in the current operating point category under the inverse covariance matrix of the current operating point category, and the larger the c is, the lower the possibility that the historical operating data of the current operating point category belongs to the current operating point category;

[0030] According to the occurrence probability of each working point category, the sum value, the discrimination value and the fourth relational expression The weight of each of the historical operating data belonging to each category is determined, wherein W is the weight of each of the historical operating data belonging to each category, and P is the occurrence probability of each of the operating point categories.

[0031] On the other hand, updating the operating point category corresponding to the historical operating data according to the weight includes:

[0032] The weights of each of the historical operating data belonging to each operating point category are combined into a weight matrix W i a weight representing that each of the historical operating data belongs to the i-th operating point category, where i is an integer not greater than N;

[0033] Each row of the weight matrix is ​​normalized, and the column where the maximum value of each row is located is determined as the working point category corresponding to the historical operating data.

[0034] On the other hand, after updating the operating point category corresponding to the historical operating data according to the weight, the method further includes:

[0035] According to the weight of each category of the historical operation data, the historical operation data and the fifth relational expression Determine a sparse matrix S for each of the working point categories, wherein the sparse matrix represents the correlation between any two data in the operating data, and the larger the absolute value of the value in the sparse matrix is, the stronger the correlation is;

[0036] Calculate a new inverse covariance matrix θ according to the sparse matrix and the sixth relation min(-log(det(θ))+tr(θ*S)+λθ1), where λ is a non-negative parameter, tr(θ*S) is the trace of θ*S, and det(θ) is the determinant of θ;

[0037] Using the new inverse covariance matrix as the inverse covariance matrix of the historical operating data;

[0038] According to the weight of each of the historical operating data belonging to each category, the occurrence probability of each operating point category, the sum value, the discrimination value and the seventh relational expression Determine the target value obj;

[0039] Where W i The weight that represents each of the historical operating data belongs to the i-th operating point category, P i Characterizing the historical operating data of the i-th operating point category, c i Characterize the discriminant value of the i-th working point category, q i Characterize the discriminant value of the i-th working point category;

[0040] If the difference between the target value and the target value obtained last time is less than the difference threshold, then return to step to determine the occurrence probability of each working point category;

[0041] If the difference between the target value and the target value obtained last time is not less than the difference threshold, the process proceeds to a step of calculating a preset sample reachable value based on the inverse covariance matrix of the historical operating data.

[0042] On the other hand, calculating a preset sample reachable value according to the inverse covariance matrix of the historical operation data includes:

[0043] min(-log e (det(θ train ))+tr(θ train *S train )+λ train ||V train||1) as the objective function, As a constraint, determine the difference image matrix θ train ;

[0044] Among them, det(θ train ) is θ train The determinant of tr(θ train *S train ) is θ train *S train The trace of train is an adjustable parameter, V train is the semi-inverse difference matrix of the difference matrix, S train =(X train -μ train ) T (X train -μ train ), S train is the sparse value of the historical operation data, representing the degree of difference between each variable of the historical operation data, μ train is the mean value of the historical operating data;

[0045] Determine the maximum eigenvalue eig of the difference image matrix train And determine the reference maximum characteristic root eig of each operating point type according to the inverse covariance matrix of the historical operating data r ;

[0046] According to the maximum characteristic root, the reference maximum characteristic root and the eighth relational expression Determine the preset sample reachable value, where A train is the preset sample reachable value, abs(eig train -eig r ) characterizes eig train -eig r The absolute value of .

[0047] On the other hand, determining the variable causing the alarm in the operation data according to the inverse covariance matrix of the current cycle includes:

[0048] Determine a semi-inverse difference map matrix of the inverse covariance matrix of the current period according to the inverse covariance matrix of the current period;

[0049] Square the elements in the semi-inverse difference matrix and then sum them row by row;

[0050] The largest variable in the vector is used as the variable causing the alarm in the operating data.

[0051] In order to solve the above technical problems, the present invention also provides a distillation tower fault determination and alarm device, comprising:

[0052] Memory for storing computer programs;

[0053] A processor is used to implement the steps of the above-mentioned method for determining a distillation tower failure when executing the computer program.

[0054] The present application provides a method and device for determining a distillation tower fault, which relates to the field of industrial process detection and control, and is applied to a distillation tower, including determining the inverse covariance matrix of the historical operating data of the distillation tower; calculating a preset sample reachable value based on the inverse covariance matrix of the historical operating data; and determining the inverse covariance matrix of the operating data of the current cycle. According to the historical operating data, an inverse covariance matrix for determining the relationship between each influencing variable and the maximum value that the historical operating data as a whole can reach, i.e., the preset sample reachable value, can be obtained. When the operating data of the current cycle is obtained, by calculating the sample reachable value of the current cycle and comparing it with the preset sample reachable value, it can be determined whether a fault currently occurs. When a fault occurs, the specific cause of the fault, i.e., the variable that causes the alarm, is determined based on the inverse covariance matrix of the operating data of the current cycle. An alarm is given in time for the occurrence of a fault, and the cause of the fault can be determined, and the fault can be accurately located. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the prior art and the drawings required for use in the embodiments 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 creative work.

[0056] Figure 1 A flow chart of a method for determining a distillation tower failure provided by the present invention;

[0057] Figure 2 A schematic structural diagram of a distillation tower provided by the present invention;

[0058] Figure 3 A schematic structural diagram of a device for determining a distillation tower fault provided by the present invention. DETAILED DESCRIPTION

[0059] The core of the present invention is to provide a method and device for determining a distillation tower fault, determine the specific cause of the fault, that is, the variable causing the alarm, give an alarm in time for the fault, determine the cause of the fault, and accurately locate the fault.

[0060] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] In the field of process industry, distillation tower is a common equipment. Taking advantage of the different boiling points of different substances, distillation tower is generally used to separate and purify mixed raw materials. The distillation tower maintains a certain temperature and pressure inside the distillation tower by heating the tower kettle, and then produces the target component at the top of the tower. In order to ensure the purity and stability of the target component, it is very important to maintain the smooth operation of the distillation tower. However, with the increase of the operation time of the production equipment, due to equipment aging, failure, impurities in the materials, etc., the distillation tower may have problems such as tray blockage and tower body damage, which will affect the operation efficiency of the distillation tower and even affect production safety. The failure of the distillation tower is generally caused by multiple variables such as temperature, pressure, flow, etc. Due to the mutual influence between these multiple variables, the relevant technology cannot accurately give the root cause variable that causes the failure of the distillation tower, that is, to find the most important variable from temperature, pressure, flow, etc., which causes the monitoring statistic to exceed the limit. Therefore, it is of great significance to establish a method that can accurately describe the interaction relationship between variables.

[0062] Figure 1 A flow chart of a method for determining a distillation tower failure provided by the present invention, Figure 2 A schematic structural diagram of a distillation tower provided by the present invention;

[0063] The method for determining a distillation tower failure is applied to a distillation tower, wherein the upper part of the distillation tower is the tower top, the lower part of the distillation tower is the tower bottom, and raw materials are input from the middle part of the distillation tower and output from the tower top and the tower bottom respectively;

[0064] The method for determining the distillation tower failure includes:

[0065] S11: determining an inverse covariance matrix of historical operating data of the distillation tower, the operating data including a combination of two or more of the pressure at the top of the tower, the temperature at the top of the tower, the temperature at the bottom of the tower, the liquid level height at the bottom of the tower, and the temperature in the middle of the distillation tower, the numerical value in the inverse covariance matrix represents the causal relationship between any two operating data in the operating data, and the numerical value is positively correlated with the causal relationship;

[0066] Since there are many reasons for the failure of the distillation tower, multiple variables may cause it, or multiple variables may affect each other, for example, variable 1 affects variable 2, and both variable 1 and variable 2 can cause the failure of the distillation tower, so it is necessary to determine the causal relationship between the variables.

[0067] The inverse covariance matrix can represent the relationship between data. For example, (i, j) represents the causal relationship between the i-th variable and the j-th variable. If the value of (i, j) is larger, the causal relationship is stronger, and the possibility of mutual influence between variables i and j is greater.

[0068] Considering that there are many influencing variables, a combination of two or more of the five variables, namely, the pressure at the top of the tower, the temperature at the top of the tower, the temperature at the bottom of the tower, the liquid level height at the bottom of the tower and the temperature in the middle of the distillation tower, are selected to determine the relationship between the variables.

[0069] The time of obtaining historical operation data can be set according to actual needs

[0070] S12: Calculating a preset sample reachable value according to an inverse covariance matrix of historical operation data, where the sample reachable value represents the maximum value of the operation data of the distillation tower;

[0071] In order to determine the point where the operating data of the distillation tower fails, the sample reachable value is determined by the inverse covariance matrix. It can be understood that the sample reachable value represents the maximum value of the operating data. If the data of the current cycle is greater than the preset sample reachable value, it proves that the operating data of the current cycle may cause the distillation tower to fail and an alarm is required.

[0072] S13: Determine the inverse covariance matrix of the operating data of the current cycle;

[0073] S14: Determine the sample reachable value of the current period according to the inverse covariance matrix of the current period;

[0074] S15: When the sample reachable value of the current period is greater than the preset sample reachable value, an alarm is issued;

[0075] Determine the inverse covariance matrix of the running data of the current cycle and determine the sample reachability of the current cycle. Whether a fault occurs can be determined by the sample reachability of the current cycle and the preset sample reachability, and an alarm is issued when a fault occurs.

[0076] S16: Determine the variable causing the alarm in the operating data according to the inverse covariance matrix of the current period.

[0077] Since the inverse covariance matrix can determine the relationship between the variables of the operating data of the current period, the relationship between the variables can be determined according to the inverse covariance matrix of the current period, and then the data that causes the alarm can be determined.

[0078] The present application provides a method for determining a distillation tower fault, which relates to the field of industrial process detection and control, and is applied to a distillation tower, including determining the inverse covariance matrix of the historical operating data of the distillation tower; calculating a preset sample reachable value based on the inverse covariance matrix of the historical operating data; and determining the inverse covariance matrix of the operating data of the current cycle. According to the historical operating data, an inverse covariance matrix for determining the relationship between each influencing variable and the maximum value that the historical operating data as a whole can reach, i.e., the preset sample reachable value, can be obtained. When the operating data of the current cycle is obtained, by calculating the sample reachable value of the current cycle and comparing it with the preset sample reachable value, it can be determined whether a fault currently occurs. When a fault occurs, the specific cause of the fault, i.e., the variable that causes the alarm, is determined based on the inverse covariance matrix of the operating data of the current cycle. An alarm is issued in a timely manner for the occurrence of a fault, and the cause of the fault can be determined, and the fault can be accurately located.

[0079] Based on the above embodiments:

[0080] In some embodiments, determining an inverse covariance matrix of historical operating data of a distillation column includes:

[0081] Determine the mean of historical operating data;

[0082] According to the historical operation data, the mean of the historical operation data and the first relationship θ=(X-μ)(X-μ) T Determine the inverse covariance matrix of the historical operating data;

[0083] Where θ is the inverse covariance matrix of the historical operating data, X is the historical operating data, and μ is the mean of the historical operating data.

[0084] First, calculate the mean of the historical operation data. For example, the operation data includes the pressure at the top of the tower, the temperature at the top of the tower, the temperature at the bottom of the tower, the liquid level of the bottom of the tower, and the temperature in the middle of the distillation tower. Then, the mean includes the mean of the pressure at the top of the tower, the temperature at the top of the tower, the temperature at the bottom of the tower, the liquid level of the bottom of the tower, and the temperature in the middle of the distillation tower. Then, use these data and the mean of the corresponding data to calculate the inverse covariance matrix.

[0085] In some embodiments, before determining the inverse covariance matrix of the historical operating data of the distillation column, the method further includes:

[0086] The historical operating data of the distillation tower is divided into N types of operating points using the k-means clustering algorithm;

[0087] Determine the probability of occurrence of each work point category;

[0088] Determine the weight of each historical operating data belonging to each category according to the occurrence probability and the inverse covariance matrix of each category of historical operating data, the weight representing the possibility that the historical operating data belongs to each operating point category;

[0089] Update the operating point category corresponding to the historical operating data according to the weight.

[0090] Considering that the distillation tower operating point changes, it is impossible to effectively adapt to the change, and then diagnose the operating point change as an unstable distillation tower system, which is obviously unreasonable. Therefore, it is necessary to use the k-means clustering algorithm to classify the historical operation data. For example, if the number of types is set to N, the algorithm will classify the historical operation data into N categories.

[0091] Considering that the classification standard of data is not accurate enough in the classification process, the classified data needs to be corrected. The weight of each set of operating data belonging to the working point category is calculated. The higher the weight, the higher the possibility that the set of data belongs to the working point category.

[0092] The weight of each set of operating data belonging to each operating point category is calculated. The higher the weight, the higher the possibility that the set of operating data belongs to the operating point category. Finally, the operating point category with the highest weight is selected as the operating point category of the set of historical operating data.

[0093] In some embodiments, after determining the inverse covariance matrix of the operation data of the current cycle, the method further includes:

[0094] Determine the weight of each category of the operating data of the current period according to the inverse covariance matrix of the operating data of the current period;

[0095] The category with the largest weight is used as the category of the running data of the current cycle;

[0096] When the sample reachable value of the current period is greater than the preset sample reachable value, an alarm is generated, including:

[0097] When the sample reachable value of the current period is greater than the preset sample reachable value corresponding to the category of the running data of the current period, an alarm is issued.

[0098] After obtaining the operating data of the current cycle, the working point type of the operating data of the current cycle is determined in the above manner, and then when calculating whether a fault occurs, the sample reachable value of the current cycle is compared with the preset sample reachable value corresponding to the current cycle to determine whether the operating data under the current working point type has a fault, thereby avoiding misjudgment of faults due to different working points.

[0099] In some embodiments, the weight of each historical operation data belonging to each category is determined according to the occurrence probability and the inverse covariance matrix of each category of historical operation data, including:

[0100] Determine the upper triangular matrix L of the inverse covariance matrix of the historical operating data;

[0101] According to the upper triangular matrix, historical operating data and the second relationship Determine a summation value q, the summation value represents that each historical operating data is not significant under the inverse covariance matrix of the current operating point category. The closer q is to 0, the more significant the historical operating data belongs to the current operating point category. X is the historical operating data;

[0102] The discriminant value c is determined according to the upper triangular matrix and the third relation c=log22π+2||log(diag(L))||1. The discriminant value represents the overall insignificance of all historical operating data in the current operating point category under the inverse covariance matrix of the current operating point category. The larger the c is, the lower the possibility that the historical operating data of the current operating point category belongs to the current operating point category.

[0103] According to the occurrence probability, sum value, discrimination value and the fourth relation of each working point category Determine the weight of each historical operating data belonging to each category, where W is the weight of each historical operating data belonging to each category, and P is the occurrence probability of each working point category.

[0104] The elements on the inverse covariance matrix reflect the degree of causal association between the variables in the row and column. The closer to 0, the more independent it is, and the larger the absolute value, the more causal correlation. The matrix is ​​a symmetric matrix, and the upper triangular matrix L is used for calculation. q represents the insignificance (greater than or equal to 0) of each sample in the historical data under the inverse covariance matrix distribution of the current category. The closer the insignificance is to 0, the more significantly the sample belongs to this category. The sum value q is a column vector with a length equal to the length of the historical data. The calculation formula means that the inverse of the matrix L is multiplied by the matrix X, and then each element in the calculation result is squared, and then each row of the matrix is ​​summed. The discriminant value c represents the overall insignificance of the data (negative or positive) under the inverse covariance matrix distribution of the current category. The greater the insignificance, the lower the real possibility that these samples under this category belong to this category. The meaning of the second term of the calculation formula is to take the logarithm (base 2) of the elements on the diagonal of the matrix L to obtain a vector, and take the 1 norm of the vector, so c is a scalar. Determine the proportion of each sample belonging to this category. This W is a vector with N rows and 1 column, where N represents the number of samples in the historical data.

[0105] In some embodiments, updating the operating point category corresponding to the historical operating data according to the weight includes:

[0106] Combine the weights of each historical operating data belonging to each operating point category into a weight matrix W i The weight that represents each historical operating data belongs to the i-th operating point category, where i is an integer not greater than N;

[0107] Each row of the weight matrix is ​​normalized, and the column where the maximum value of each row is located is determined as the working point category corresponding to the historical operating data.

[0108] After processing the samples of N categories, the weight vectors W of N categories are merged into a weight matrix Where W i Represents the W calculated for each category. Normalize each row of , and select the column with the maximum value in each row as the category to which the sample belongs.

[0109] In some embodiments, after updating the operating point category corresponding to the historical operating data according to the weight, the method further includes:

[0110] According to the weight of each category of each historical operation data, the historical operation data and the fifth relation Determine a sparse matrix S for each type of working point. The sparse matrix represents the correlation between any two data in the operating data. The larger the absolute value of the value in the sparse matrix, the stronger the correlation.

[0111] Calculate the new inverse covariance matrix θ according to the sparse matrix and the sixth relation min(-log(det(θ))+tr(θ*S)+λθ1), where λ is a non-negative parameter, tr(θ*S) is the trace of θ*S, and det(θ) is the determinant of θ;

[0112] The new inverse covariance matrix is ​​used as the inverse covariance matrix of the historical operating data;

[0113] According to the weight of each category of each historical operating data, the probability of occurrence of each working point category, the sum value, the discrimination value and the seventh relationship Determine the target value obj;

[0114] Where W i The weight that represents each historical operating data belongs to the i-th operating point category, P i Characterize the historical operating data of the i-th operating point category, c i Characterize the discriminant value of the i-th working point category, q i Characterize the discriminant value of the i-th working point category;

[0115] If the difference between the target value and the last target value is less than the difference threshold, then return to step 1 to determine the occurrence probability of each working point category;

[0116] If the difference between the target value and the target value obtained last time is not less than the difference threshold, the process proceeds to a step of calculating a preset sample reachable value based on the inverse covariance matrix of the historical operation data.

[0117] The sparse matrix represents the covariance relationship between variables after weight correction. The larger the absolute value of the elements in the sparse matrix, the more obvious the correlation between the two variables. The method of updating the inverse covariance matrix of all categories based on the sparse matrix of all categories. After updating the inverse covariance matrix, the target value is calculated. The target value is used to determine whether to continue the correction or proceed to the next step.

[0118] In some embodiments, calculating a preset sample reachable value according to an inverse covariance matrix of historical operation data includes:

[0119] min(-log e (det(θ train ))+tr(θ train *S train )+λ train ||V train ||1) as the objective function, As a constraint, determine the difference image matrix θ train ;

[0120] Among them, det(θ train ) is θ train The determinant of tr(θ train *S train ) is θ train *S train The trace of train is an adjustable parameter, V train is the semi-inverse difference matrix of the difference matrix, S train =(X train -μ train ) T (X train -μ train ), S train is the sparse value of the historical operation data, representing the degree of difference between each variable of the historical operation data, μ train is the mean of historical operating data;

[0121] Determine the maximum eigenvalue eig of the difference matrix train And determine the reference maximum characteristic root eig for each operating point type based on the inverse covariance matrix of historical operating datar ;

[0122] According to the maximum characteristic root, the reference maximum characteristic root and the eighth relation Determine the preset sample reachable value, where A train is the preset sample reachable value, abs(eig train -eig r ) characterizes eig train -eig r The absolute value of .

[0123] In order to calculate an indicator as a threshold for stability judgment, for each sample, the difference map matrix is ​​calculated according to the inverse covariance matrix and reference mean of the category to which it belongs. The difference map matrix is ​​an inverse covariance matrix. It is based on the original inverse covariance matrix of the category, with the goal of maximizing the uniformity of the causal relationship between single sample variables. The matrix result is solved, the maximum eigenvalue of the obtained difference map matrix is ​​calculated, and the sample reachable value is calculated in combination with the reference maximum eigenvalue of the category to represent the indicator that the sample can achieve, that is, how much percentage it can exceed the reference maximum eigenvalue.

[0124] In some embodiments, determining the variable causing the alarm in the operating data according to the inverse covariance matrix of the current period includes:

[0125] Determine a semi-inverse difference map matrix of the inverse covariance matrix of the current period according to the inverse covariance matrix of the current period;

[0126] Square the elements in the semi-inverse difference matrix and sum them row by row;

[0127] The largest variable in the vector is taken as the variable causing the alarm in the running data.

[0128] For samples that issue an alarm, the root cause probability of each variable in the sample is calculated as follows:

[0129] For V obtained in step 1 of the online phase test Square the elements and then sum them row by row to get a column vector with a length equal to the number of variables. The variable corresponding to the largest element in the vector is the root cause of this alarm.

[0130] From the perspective of the formula, V test It is a 5*5 matrix, and after calculation, a 5*1 vector is obtained.

[0131] Figure 3 A schematic diagram of the structure of a device for determining a distillation tower fault provided by the present invention, the device comprising:

[0132] Memory for storing computer programs;

[0133] The processor is used to implement the steps of the method for determining the fault of the upper distillation tower when executing the computer program.

[0134] For an introduction to the device for determining a distillation tower failure provided in the present application, please refer to the above-mentioned embodiments, which will not be described in detail here.

[0135] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0136] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0137] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining a distillation tower failure, characterized in that: Applied to a distillation tower, the upper part of the distillation tower is the tower top, the lower part of the distillation tower is the tower bottom, the raw materials are input from the middle part of the distillation tower, and are output from the tower top and the tower bottom respectively; The method for determining the distillation tower failure comprises: Determine an inverse covariance matrix of historical operating data of the distillation tower, wherein the operating data includes a combination of two or more of the pressure at the tower top, the temperature at the tower top, the temperature at the tower bottom, the liquid level height at the tower bottom, and the temperature in the middle of the distillation tower, wherein a numerical value in the inverse covariance matrix represents a causal relationship between any two operating data in the operating data, and the numerical value is positively correlated with the causal relationship; Calculating a preset sample reachable value according to an inverse covariance matrix of the historical operation data, wherein the sample reachable value represents a maximum value of the operation data of the distillation tower; Determine the inverse covariance matrix of the running data for the current period; Determine the sample reachable value of the current period according to the inverse covariance matrix of the current period; When the sample reachable value of the current period is greater than the preset sample reachable value, an alarm is issued; Determining the variable causing the alarm in the operation data according to the inverse covariance matrix of the current cycle; Calculating a preset sample reachable value according to an inverse covariance matrix of the historical operation data includes: min(-log e (det(θ train ))+tr(θ train *S train )+λ train ||V train ||1) as the objective function, As a constraint, determine the difference image matrix θ train ; Among them, det(θ train ) is θ train The determinant of tr(θ train *S train ) is θ train *S train The trace of train is an adjustable parameter, V train is the semi-inverse difference matrix of the difference matrix, S train =(X train -μ train ) T (X train -μ train ), S train is the sparse value of the sample data in the historical operation data, representing the degree of difference between each variable of the sample data in the historical operation data, μ train is the mean of the sample data in the historical operation data, θ is the inverse covariance matrix of the historical operation data, X train is sample data in the historical operation data; Determine the maximum eigenvalue eig of the difference image matrix train And determine the reference maximum characteristic root eig of each operating point type according to the inverse covariance matrix of the historical operating data r ; According to the maximum characteristic root, the reference maximum characteristic root and the eighth relational expression Determine the preset sample reachable value, where A train is the preset sample reachable value, abs(eig train -eig r ) characterizes eig train -eig r The absolute value of .

2. The method for determining a distillation tower failure according to claim 1, wherein: Determining an inverse covariance matrix of the historical operating data of the distillation column includes: Determining a mean of the historical operating data; According to the historical operation data, the mean of the historical operation data and the first relational expression θ=(X-μ)(X-μ) T determining an inverse covariance matrix of the historical operating data; Wherein θ is the inverse covariance matrix of the historical operating data, X is the historical operating data, and μ is the mean of the historical operating data.

3. The method for determining a distillation tower failure according to claim 1, wherein: Before determining the inverse covariance matrix of the historical operating data of the distillation tower, the method further includes: Using a k-means clustering algorithm to classify the historical operating data of the distillation tower into N types of operating points; Determine the probability of occurrence of each work point category; Determine a weight of each of the historical operating data belonging to each category according to the occurrence probability and the inverse covariance matrix of the historical operating data of each operating point category, wherein the weight represents the possibility that the historical operating data belongs to each operating point category; The operating point category corresponding to the historical operating data is updated according to the weight.

4. The method for determining a distillation tower failure according to claim 3, wherein: After determining the inverse covariance matrix of the running data of the current period, it also includes: Determine the weight of each category of the operation data of the current period according to the inverse covariance matrix of the operation data of the current period; Using the category with the largest weight as the category of the operation data of the current cycle; When the sample reachable value of the current period is greater than the preset sample reachable value, an alarm is issued, including: An alarm is issued when the sample reachable value of the current period is greater than the preset sample reachable value corresponding to the category of the running data of the current period.

5. The method for determining a distillation tower failure according to claim 3, wherein: Determining the weight of each of the historical operating data belonging to each category according to the occurrence probability and the inverse covariance matrix of the historical operating data of each operating point category includes: Determine an upper triangular matrix L of the inverse covariance matrix of the historical operating data; According to the upper triangular matrix, the historical operation data and the second relational expression Determine a summation value q, wherein the summation value represents the insignificance of each of the historical operating data under the inverse covariance matrix of the current operating point category, and the closer q is to 0, the more significantly the historical operating data belongs to the current operating point category, and X is the historical operating data; Determine a discriminant value c according to the upper triangular matrix and the third relational expression c=log22π+2||log(diag(L))||1, wherein the discriminant value represents the overall insignificance of all the historical operating data in the current operating point category under the inverse covariance matrix of the current operating point category, and the larger the c is, the lower the possibility that the historical operating data of the current operating point category belongs to the current operating point category; According to the occurrence probability of each working point category, the sum value, the discrimination value and the fourth relational expression The weight of each of the historical operating data belonging to each category is determined, wherein W is the weight of each of the historical operating data belonging to each category, and P is the occurrence probability of each of the operating point categories.

6. The method for determining a distillation tower failure according to claim 5, wherein: Updating the operating point category corresponding to the historical operating data according to the weight includes: The weights of each of the historical operating data belonging to each operating point category are combined into a weight matrix W i a weight representing that each of the historical operating data belongs to the i-th operating point category, where i is an integer not greater than N; Each row of the weight matrix is ​​normalized, and the column where the maximum value of each row is located is determined as the working point category corresponding to the historical operating data.

7. The method for determining a distillation tower failure according to claim 5, wherein: After updating the operating point category corresponding to the historical operating data according to the weight, the method further includes: According to the weight of each category of the historical operation data, the historical operation data and the fifth relational expression Determine a sparse matrix S for each of the working point categories, wherein the sparse matrix represents the correlation between any two data in the operating data, and the larger the absolute value of the value in the sparse matrix, the stronger the correlation; Calculate a new inverse covariance matrix θ according to the sparse matrix and the sixth relation min(-log(det(θ))+tr(θ*S)+λθ1), where λ is a non-negative parameter, tr(θ*S) is the trace of θ*S, and det(θ) is the determinant of θ; Using the new inverse covariance matrix as the inverse covariance matrix of the historical operating data; According to the weight of each of the historical operating data belonging to each category, the occurrence probability of each operating point category, the sum value, the discrimination value and the seventh relational expression Determine the target value obj; Where W i The weight that represents each of the historical operating data belongs to the i-th operating point category, P i Characterizing the historical operating data of the i-th operating point category, c i Characterize the discriminant value of the i-th working point category, q i Characterize the discriminant value of the i-th working point category; If the difference between the target value and the target value obtained last time is less than the difference threshold, then return to step to determine the occurrence probability of each working point category; If the difference between the target value and the target value obtained last time is not less than the difference threshold, the process proceeds to a step of calculating a preset sample reachable value based on the inverse covariance matrix of the historical operating data.

8. The method for determining a distillation tower failure according to claim 1, wherein: Determining the variable causing the alarm in the operation data according to the inverse covariance matrix of the current cycle includes: Determine a semi-inverse difference map matrix of the inverse covariance matrix of the current period according to the inverse covariance matrix of the current period; Square the elements in the semi-inverse difference matrix and then sum them row by row; The largest variable in the vector is used as the variable causing the alarm in the operating data.

9. A distillation tower fault determination and alarm device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the method for determining a distillation tower failure as claimed in any one of claims 1 to 8 when executing the computer program.

Images