Industrial multivariable alarm method and system based on search cone feasible working domain modeling

Abstract

The present invention provides an industrial multivariable alarm method and system based on search cone feasible working domain modeling. The method determines associated variables determined based on trend changes as monitoring variables. Data points of the acquired monitoring variables are preprocessed. A search cone is searched in a feasible working domain boundary model pre-constructed based on historical monitoring variable data to determine whether the preprocessed data points can be included. If so, the search cone is considered a normal data point; otherwise, it is an abnormal data point. The normal data point corresponding to the abnormal data point is determined. A straight line corresponding to a dynamic alarm threshold is calculated, and the intersection of the straight line and the feasible working domain boundary model is calculated to determine the dynamic alarm threshold. If the preprocessed data point is outside the feasible working domain boundary model, or if a variable exceeds the operating range represented by its corresponding alarm threshold, an alarm is triggered. The present invention can implement the design of multivariable alarm thresholds in modern industrial systems, and the design is reasonable and has good versatility.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial alarms, and in particular relates to an industrial multivariable alarm method and system based on search cone feasible working domain modeling. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the increasing automation of industrial production processes, factories are deploying alarm systems for an increasing number of variables. However, inadequate alarm system design can lead to excessive nuisance alarms, preventing operators from identifying true alarms and delaying the timely resolution of abnormal conditions. Furthermore, because variables in a system often correlate with each other, alarm systems designed for a single variable fail to account for the interrelationships between them. This results in a failure to detect true anomalies in these correlated variables, leading to numerous nuisance alarms. This can worsen the anomaly, causing economic losses and casualties. Therefore, it is necessary to consider the design of multivariable alarm systems to address the problem of excessive nuisance alarms.

[0004] The inventors understand that currently, to address the problem of excessive nuisance alarms in multivariable alarm systems, methods such as multivariate statistics, transfer entropy, Bayesian filters, and alarm cluster analysis have been developed. However, in actual industrial production processes, due to the movement of materials and the continuous conversion of energy, numerous process variables are involved. These monitored variables are closely interrelated and often have some degree of connection.

[0005] Existing multivariable alarm systems often have the following problems:

[0006] 1) The relationship between the monitored variables is not fully utilized to design the alarm threshold, resulting in an unreasonable design of the alarm threshold; 2) The multivariable alarm system design method based on the hyperellipsoid to establish a feasible working domain can only be used to represent some process variables that completely conform to the Gaussian distribution in any dimensional space, and is not very universal; 3) The multivariable alarm system design method based on the convex hull model to establish a feasible working domain has a complex calculation process, consumes too much computing resources during the solution optimization calculation, and is prone to underfitting. The higher the dimension, the more serious the situation is, and there may even be no solution, which lacks practicality. Summary of the Invention

[0007] In order to solve the above problems, the present invention proposes an industrial multivariable alarm method and system based on search cone feasible working domain modeling. The present invention can realize the design of multivariable alarm thresholds in modern industrial systems. The design is reasonable and has good versatility.

[0008] It should be pointed out that the industrial multivariable alarm method / system provided by the present invention can be applied to equipment involved in industrial production, such as three-capacity water tanks, transformers, etc., and can also be applied to large systems involved in industrial production, such as coal grinding machine pulverizing systems, etc. It can also be applied to detection and alarm in industrial production environments, such as fire detection, gas leak detection, etc.; the alarm system or design method used in the above-mentioned application objects or application scenarios, if directly applied or simply transformed from the technical solution provided by the present invention, should fall within the scope of protection of the present invention.

[0009] According to some embodiments, the present invention adopts the following technical solutions:

[0010] An industrial multivariable alarm method based on search cone feasible working domain modeling includes the following steps:

[0011] Determine the associated variables determined based on trend changes as monitoring variables;

[0012] The data points of the acquired monitoring variables are pre-processed, and a search cone is found in a feasible working domain boundary model pre-built based on the historical monitoring variable data to see whether there is a search cone that can contain the pre-processed data points. If so, it is a normal data point, otherwise it is an abnormal data point;

[0013] Determine the normal data points corresponding to the abnormal data points;

[0014] Based on the normal data points, calculate the straight line corresponding to the dynamic alarm threshold, calculate the intersection of the straight line and the feasible working area boundary model, and then determine the dynamic alarm threshold;

[0015] If the preprocessed data point is outside the feasible working domain boundary model, or if a variable exceeds the operating range represented by its corresponding alarm threshold, an alarm is triggered.

[0016] As an optional implementation method, the specific process of determining the associated variables determined based on trend changes as monitoring variables includes: using a bottom-up piecewise linear representation method to perform piecewise linear representation on the sample data, obtaining the trend change relationship of each variable, and then determining the monitoring variable based on the trend change relationship and the actual relationship between each variable.

[0017] As an optional implementation, the specific process of pre-building a feasible work domain boundary model based on historical monitoring variable data includes:

[0018] Preprocess the normal data points of the monitoring variables in the historical monitoring data, regard the feasible working domain composed of the monitoring variables as consisting of multiple search cones, and determine the side edge length of the search cone based on the preprocessed data;

[0019] Set the optimal step angle of the search cone, select the values ​​of the corner coordinates of each search cone, and obtain the vertex coordinates of the infinitesimal element according to the arrangement and combination of all corner coordinates, and form a vertex set together with the origin;

[0020] According to the vertex set, the search cone expression is determined, and the feasible working domain boundary model is obtained through traversal search.

[0021] As a further limitation, the specific process of preprocessing includes: representing the historical normal data points of each monitoring variable using a vector, and standardizing the data of each monitoring variable using a Z-score standardization method.

[0022] As a further limitation, the search cone is represented by a convex hull in a high-dimensional space, and the vertex set of the convex hull is composed of the origin and the vertices of the hypercube at the bottom of the search cone.

[0023] As a further qualification, the specific process of setting the optimal search cone step angle includes:

[0024] The vertex angle between two adjacent edges of the search cone is defined as the step angle α of the search cone. The value of α is determined by the minimum value of the following objective function F(α):

[0025] F(α)=(1-η)σ c ε r

[0026] Among them, η represents the fitness index of the feasible working domain boundary model; σ c Represents the comprehensive standard deviation of the alarm thresholds of adjacent monitoring data points; ε r Indicates the radius difference index of adjacent search cones.

[0027] As a further limitation, the values ​​of the angular coordinates of each edge of the search cone are selected separately. The specific process of obtaining the coordinates of the vertex of the infinitesimal element according to the arrangement and combination of all angular coordinates includes: setting each angular coordinate equal to 0 and the step angle α, searching the angular coordinates of each edge of the cone in n-dimensional space [φ1, φ2,…, φ n-1 ]Total 2 n-1 The different permutations and combinations correspond to the 2 n-1 Vertices, according to the arrangement and combination of all angular coordinates, the coordinates of the infinitesimal vertex are obtained.

[0028] As a further limitation, based on the vertex set, the specific process of determining the search cone expression includes: finding the data points on all hyperplanes of the convex hull through the fast convex hull algorithm, using the expression to represent the i-th hyperplane that forms the convex hull, indexing the n data points on the i-th hyperplane one by one and substituting them into the expression, solving to obtain the normal vector and hyperplane offset, calculating all hyperplanes that form the convex hull one by one, obtaining the complete expression of the hyperplane to represent the search cone, and obtaining the search cone expression.

[0029] As an optional implementation method, the straight line corresponding to the dynamic alarm threshold is calculated, and the intersection of the straight line and the feasible working domain boundary model is calculated, and then the specific process of determining the dynamic alarm threshold includes: according to the straight line expression corresponding to the dynamic alarm threshold of a certain monitoring variable, the straight line expression is linked with the feasible working domain boundary model to obtain the intersection of the two, forming a set Z j , according to the set Z j The maximum and minimum values ​​determine the upper and lower limits of the dynamic alarm threshold.

[0030] An industrial multivariable alarm system based on search cone feasible working domain modeling, comprising:

[0031] a monitoring variable determination module configured to determine the associated variable determined based on the trend change as a monitoring variable;

[0032] The data point judgment module is configured to pre-process the data points of the acquired monitoring variables and search whether a search cone can contain the pre-processed data points in a feasible working domain boundary model pre-built based on historical monitoring variable data. If so, the search cone is a normal data point, otherwise it is an abnormal data point;

[0033] A dynamic alarm threshold determination module is configured to determine normal data points corresponding to abnormal data points, calculate a straight line corresponding to the dynamic alarm threshold based on the normal data points, calculate the intersection of the straight line and the feasible working area boundary model, and then determine the dynamic alarm threshold;

[0034] The alarm module is configured to trigger an alarm if the preprocessed data point is outside the feasible working domain boundary model, or if a certain variable exceeds the operating range represented by its corresponding alarm threshold, otherwise no alarm is triggered.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention calculates abnormal data points by judging whether they are within the search cone model, thereby simplifying the calculation process and improving the practicability of the model; it overcomes the problem of abnormal data points existing in the alarm method based on the convex hull model, excessive consumption of computing resources during solution optimization calculation, the higher the dimension, the more serious the situation, and even the possibility of no solution, which is not conducive to practical application and has the problem of underfitting.

[0037] The present invention adopts a method of preprocessing data to achieve data screening, fully utilizing the direct connection between the data of each variable, and only requires the historical normal data points of the process variables to form a closed area of ​​the convex hull model in the high-dimensional space. It overcomes the shortcomings of the hyperellipsoid feasible working domain boundary model that the process variables are limited to satisfy the Gaussian distribution and the convex hull model that the historical normal data points of the process variables must form a convex closed area in the high-dimensional space, thereby increasing versatility.

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0040] Figure 1 A flow chart of a multivariate alarm method based on search cone feasible working domain modeling according to at least one embodiment of the present invention;

[0041] Figure 2 The various indicators and objective functions of the feasible working range boundary model of the three-capacity water tank in Example 1;

[0042] Figure 3 The feasible working range boundary model of the three-capacity water tank in Example 1 and its fitting situation;

[0043] Figure 4 The operating status of the monitoring variable data points of the three-capacity water tank in Example 1 and its dynamic alarm line;

[0044] Figure 5 This is a trend diagram of some process variables of the coal mill in Example 2;

[0045] Figure 6 This is a trend diagram of process variables of another part of the coal mill in Example 2;

[0046] Figure 7 The original normal data of the coal mill monitoring variables in Example 2;

[0047] Figure 8 The various indicators and objective functions of the feasible working range boundary model of the coal mill in Example 2;

[0048] Figure 9 This is a diagram showing the fitting of the feasible working range boundary model based on the search cone to the historical normal data points of the coal mill in Example 2;

[0049] Figure 10 This is a diagram of the monitored operating values ​​of the process variables and their dynamic alarm thresholds in Example 2. DETAILED DESCRIPTION

[0050] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0052] like Figure 1 As shown, the present invention proposes a multivariable alarm method or alarm system design method based on search cone feasible working domain modeling, comprising the following steps:

[0053] Step 1: Preprocess the basic monitoring variables and determine the associated variables based on trend changes as the monitoring variables of the multivariable alarm system. Use the bottom-up piecewise linear representation (PLR) method to perform piecewise linear representation on the sample data to obtain the trend change relationship of each variable. Then, based on the trend change relationship and the actual mechanism relationship between each variable, the basic monitoring variables with monitoring value are obtained.

[0054] Step 2: Establish a feasible working domain boundary model based on the search cone model.

[0055] Obtain all associated process variables that the feasible working domain should contain by the method described in step 1 in represents the nth associated variable, and all historical normal data points contained in the normal working state of the production process are expressed as formula (1):

[0056]

[0057] In formula (1), represents all historical normal data points; t represents the sampling time of all historical normal data points, which can be 1s in some embodiments; N represents the number of all historical normal data points; represents the sampled data of a multivariate process at a certain moment. Then, the specific steps to establish a feasible working range boundary model based on the search cone model based on the preprocessed data obtained in the previous step are as follows:

[0058] Step 2.1: Use the Z-score standardization method to preprocess the original data of the relevant variables and standardize the original data into normalized data with a mean of 0 and a variance of 1. , the corresponding standardized data set should be:

[0059] X(t)=[x1(t),x2(t),…,x n (t)] (2)

[0060] In formula (2), x j The calculation formula for (t) is:

[0061]

[0062] In formula (3), represents the mean of the original data, Represents the standard deviation of the original data, and n represents the nth original data.

[0063] Step 2.2: All normal data points X(t) in n-dimensional space can be enclosed by multiple n-dimensional pyramids with side lengths r. These n-dimensional pyramids (with equal side lengths) are called search cones (in two-dimensional space, they are isosceles triangles with side lengths r; in three-dimensional space, they are regular square pyramids with side lengths r). Each search cone can be represented by a convex hull in a higher-dimensional space, whose vertices are the origin O and the vertices of the hypercube at the base of the search cone. Based on the concept of infinitesimals, the hypercube at the base of the search cone can be represented as Δs = Δ1Δ2…Δ n-1 , a total of 2 n-1 vertices and 2 n-1 Strip edge. Among them Δ j represents the edge of the hypercube at the base of the search cone in the Cartesian coordinate system. The feasible working domain is composed of countless tiny search cones. Based on the standardized data X(t) falling within a search cone, using Equations (4) and (5), we can obtain the side edge length r. For ease of calculation, we can take the upward modulus value of the data point in X(t) farthest from the origin O as r.

[0064]

[0065] In formula (4), X(t) represents the normalized data point, Indicates rounding up the operand and returning the smallest integer greater than the operand. |·| indicates taking the modulus value of a vector represented by a process variable. For an n-dimensional vector, the calculation formula is:

[0066]

[0067] Step 2.3: Define the vertex angle of two adjacent edges of the search cone as the step angle α of the search cone, and then let the coordinates of each angle φ be j ,j=1,2,…,n-1 is equal to 0 and α. The value of α is determined by formula (6)

[0068] It is determined when the objective function F(α) takes the minimum value.

[0069] F(α)=(1-η)σc ε r (6)

[0070] In formula (6), η represents the fitness index of the feasible working domain boundary model, and its calculation formula is shown in formula (7); σ c represents the comprehensive standard deviation of the alarm thresholds of adjacent monitoring data points, and its calculation formula is shown in formula (13); r It represents the radius difference index of adjacent search cones, and its calculation formula is shown in formula (17).

[0071]

[0072] The fitness index η in formula (7) is used to measure the fitting of the feasible working domain boundary model to the original feasible working domain. For a certain feasible working domain composed of historical normal data points, its model fitness index is defined as the proportion of data points X(t) that fill the feasible working domain boundary model. For a certain multivariable X, each process variable x j The value range of It can be expressed as the operating domain in this dimension, and the calculation expression is:

[0073]

[0074] In formula (8), min(x j ) and max(x j ) represent the minimum and maximum values ​​of all historical normal data of process variables. Then calculate each process variable x j , j=1,2,…,n, the entire feasible working domain boundary model can be divided into many equal grids (represented as superbodies in high-dimensional space), and each grid can be represented as:

[0075]

[0076] In formula (9), n represents both the number of process variables and the dimension of the superbody grid; k j Indicates the kth j The value range of a grid can be calculated by the following formula:

[0077]

[0078] In formula (10), c j (k j ) represents the minimum value interval of the jth process variable. This needs to be given by the prior knowledge of the on-site operator or reasonably determined by historical data, depending on the actual needs of the operator. The center coordinates of each super-body grid can be obtained from the above grid expression, namely:

[0079] C(k1,k2,…,k n )=[c1(k1),c2(k2),…,c n (k n )] (11)

[0080] In formula (11), c j (k j ) represents the coordinate of the superbody grid in the jth dimension, and the expression is If the center coordinates of a super-body grid are located within the boundary model of the feasible working domain, that is, the center coordinates satisfy a certain formula (25) in the boundary model of the feasible working domain, then this super-body grid is defined as an internal grid. j By changing the dimensions of each process variable, the number of all internal grids n can be obtained i For an internal superbody grid, if at least one historical normal data point falls within the superbody grid (k′1, k′2, …, k′ n ), expressed as:

[0081]

[0082] Then, the current superbody grid is defined as a valid counting grid, and all n i All internal grids are checked and calculated, and finally n c A valid counting grid.

[0083] σ c =σ h +σ l (13)

[0084] In formula (13), σ h Represents the comprehensive standard deviation index of the high alarm threshold of adjacent monitoring data points, σ l The comprehensive standard deviation index of the low alarm threshold of adjacent monitoring data points can be calculated by formula (14) and formula (15) respectively:

[0085]

[0086]

[0087] In formula (14) and formula (15), N i Indicates the number of data points for testing the feasible working domain boundary model, f i,h ,f i,l They represent the amplitude fluctuations of the adjacent data points of the i-th test data point x relative to the high alarm threshold and the low alarm threshold of the reference data point, respectively, and can be calculated by formula (16):

[0088]

[0089] In formula (16), h x ,l x They represent the high alarm threshold and low alarm threshold of the test data point x, respectively. i (t),l i (t) represent the high alarm threshold and low alarm threshold of the i-th adjacent data point of the test data point x, respectively.

[0090]

[0091] In formula (17), N r represents the number of search cones used to establish the feasible working domain boundary model, r i Represents the side edge length of the i-th search cone.

[0092] Step 2.4: Each search cone can be represented by a convex hull in a high-dimensional space. The vertex set of the convex hull is composed of the origin O and the vertices of the hypercube at the bottom of the search cone. The angular coordinates of the edges of the search cone in n-dimensional space are [φ1, φ2, …, φ n-1 ]Total 2 n-1 The different permutations and combinations in the search for the cone bottom element Δs=Δ1Δ2…Δ n-1 2 n-1 Vertices, all the permutations and combinations of angular coordinates are substituted into formula (18) one by one to obtain the coordinates of the infinitesimal vertex, and together with the origin O, they form the vertex set V of the n-dimensional space search cone.

[0093]

[0094] Step 2.5: Based on the vertex set V in step 2.4, use the fast convex hull algorithm to find the data points on all hyperplanes of the convex hull. In the high-dimensional space, the i-th hyperplane p that encloses the convex hull is (i) The general expression is:

[0095]

[0096] In formula (19), X p (t) represents the data point on the hyperplane, that is, the point on the surface satisfies the above formula. Therefore, the i-th hyperplane p that forms the convex hull is (i) It will pass through the data points on n planes, which are recorded as X(I i,1 ),X(I i,2 ),…,X(I i,n ). These data points are the data points calculated by the fast convex hull algorithm from the search cone vertex set V. Indexing the n data points on the i-th hyperplane one by one and substituting them into formula (19) yields formula (20):

[0097] a (i) XT (I i,j )-b (i) =0,j=1,2,…,n (20)

[0098] In formula (20), a (i) is the hyperplane p (i) The unit normal vector of . Then, the normal vector can be solved by combining equations (21) and (22) Offset b from the hyperplane (i) .

[0099]

[0100]

[0101] Then, let i = 1, 2, ..., m respectively to solve all the hyperplanes that form the convex hull one by one, thereby obtaining the complete expression (23) of the hyperplane, that is, the search cone expressed by formula (23).

[0102] AV′-B≤0 (23)

[0103] In formula (23), V′ is obtained by searching the cone vertex set V using the Quick-hull algorithm, and it is located on each hyperplane of the convex hull. In addition, A and B are respectively:

[0104]

[0105] In formula (24), m is the number of hyperplanes that make up the search cone in the high-dimensional space represented by the convex hull (in two-dimensional space, the hyperplane degenerates into a straight line). k∈{1,2,…,m} represents the unit normal vector of the kth hyperplane in the convex hull, b (k) ,k∈{1,2,…,m} represents the offset distance from the kth hyperplane in the convex hull to the origin O.

[0106] Step 2.6: Find the data point X in the current search cone according to formula (23) based on all historical normal data points s , and X s The data point farthest from the origin O is defined as the boundary data point X s,b , then the partial feasible working domain belonging to the current search cone can be expressed by formula (25), and the expression that can describe the current partial feasible working domain is temporarily stored in the set M.

[0107]

[0108] Step 2.7: Set the n-1th angle coordinate φ n-1 Step α. When two adjacent angular coordinates [φ1, φ2,…, φ n-1] is less than its upper limit π (or 180°), and the angular coordinates [φ1,φ2,…,φ n-2 ] If the deflection angle is less than its upper limit 2π (or 360°), then jump to step 2.4 to continue the calculation; otherwise, let φ j are 0 and α respectively, and the offset angle φ j-1 The value of is stepped by α, and the traversal process ends when φ1 does not meet the constraint.

[0109] Step 2.8: Obtain the feasible working area boundary model M according to the above calculation sequence.

[0110] Step 3: Calculate the dynamic alarm threshold and determine the alarm status.

[0111] Step 3.1: For a new monitoring data point , according to formula (3), the original monitoring data points Preprocessing is to normalize the data points X(t0).

[0112] Step 3.2: Determine whether the standardized data point X(t0) is located inside the feasible working domain boundary model: According to formula (23), find a search cone that meets the conditions in the feasible working domain boundary model. If there is a search cone that can include the data point X(t0), then the data point is a normal data point, and proceed to the next step 3.3 to calculate the dynamic alarm threshold; if there is no search cone that can include the data point X(t0), then the data point is an abnormal data point, and proceed to step 3.6 to calculate the dynamic alarm threshold.

[0113] Step 3.3: Calculate x using equations (26) and (28) j (t0) Line Y corresponding to the dynamic alarm threshold j , and then the straight line Y j The intersection of the two can be solved by linking them with the boundary model of the feasible working area and temporarily stored in the set Z j middle.

[0114]

[0115] In formula (26), p1 and p2 represent the straight line Y j Two points on , as shown in formula (27).

[0116]

[0117] C′D T +E=0 (28)

[0118] In formula (28), D represents the straight line Y j Any data point on represents the unit normal vector of the i-th hyperplane in the hyperplane cluster H, e (i) represents the offset distance from the i-th hyperplane in the hyperplane cluster H to the origin O, as shown in formula (29).

[0119]

[0120] Step 3.4: The current process variable x can be obtained by solving equation (30): j Normalized dynamic alarm threshold h of (t0) j (t0) and l j (t0), and then the dynamic alarm threshold of the original data can be obtained by formula (31) as well as

[0121]

[0122]

[0123] Step 3.5: Let j = 1, 2, ..., n respectively. All process variables x at the data point X(t0) can be solved by steps 3.3 and 3.4. j Dynamic alarm threshold.

[0124] Step 3.6: If the data point X(t0) is an abnormal data point, calculate the straight line Y passing through the data point X(t0) and the origin O using equations (26), (28) and (32): X , and solve it together with equation (33) to get the normal data point X corresponding to X(t0) M , then go to step 3.3 and calculate according to the dynamic alarm line calculation method of normal number data points.

[0125]

[0126] In formula (32), Y j,d Represents straight line Y j The unit direction vector of the hyperplane cluster H, C′ is the unit normal vector of the hyperplane cluster H and the straight line Y j The unit direction vector Y j,d n-1 different vectors that are perpendicular to each other.

[0127]

[0128] Step 3.7: Arrange the dynamic alarm thresholds of all process variables X(t0).

[0129] Step 3.8: Get all process variables After the dynamic alarm threshold is reached, the alarm variable X is judged. a Whether to send out an alarm signal, that is, when If X is located within the boundary model of the feasible working region, or each process variable is within the operating range represented by its corresponding alarm threshold, then a If the value is "0", no alarm will be triggered; If the process variable is outside the feasible working domain boundary model, or a process variable exceeds the operating range represented by its corresponding alarm threshold, then X a The value is "1", which triggers the alarm. As shown in formula (34):

[0130]

[0131] In order to make those skilled in the art more familiar with the implementation process of the method proposed in the present invention, the following is a description of specific embodiments. However, it should be noted that the scope of protection of the present invention is not limited to the following embodiments.

[0132] Example 1

[0133] The implementation of the proposed method is as follows, taking a three-tank water tank as an example. First, the proposed method is used to establish a feasible working domain boundary model, where the search cone step angle α can be determined by Figure 2 The three indicators η, σ shown c ,ε r And the objective function F(α) is optimized, Figure 2 As can be seen from (d) in the figure, the objective function achieves the minimum value when the search cone step angle is 1.4 degrees, while the other three indicators η, σ c ,ε r The values ​​of are 0.996, 0.0286 and 6.6753 respectively. The calculation results of the feasible working range boundary model are as follows: Figure 3 shown.

[0134] Calculate the dynamic alarm threshold of the test data point based on the established feasible working domain boundary model. Figure 4 In (a), the monitoring variables are designed to first run in the normal state, that is, within the feasible working domain, then gradually run outside the feasible working domain, and finally return to the feasible working domain after appropriate adjustments. Figure 4 Figures (b) and (c) show the dynamic alarm thresholds for the three-tank monitoring variables h1 and h2, respectively. The figure shows that at 1371 seconds, the level monitoring variable h2 crosses the upper limit of the dynamic alarm threshold, entering the infeasible operating range. However, at this time, the level monitoring variable h1 remains within the dynamic alarm threshold. Clearly, this multivariable system should trigger an alarm. However, monitoring only variable h1 or using static alarm lines to determine whether an alarm has occurred can easily lead to false alarms and missed alarms.

[0135] Example 2

[0136] Taking the 2018 data of the A mill of Unit 3 of Huaneng Dezhou Power Plant as an example, according to the analysis of the industrial operation data of the coal mill, the relationship between the exhaust motor current, the mill motor current, the negative pressure at the inlet of the pulverizer, the negative pressure at the inlet of the coal mill, the differential pressure at the inlet and outlet of the coal mill, the coal feed rate, the opening of the hot air door of the coal mill, the opening of the circulating air door of the coal mill, the outlet temperature of the coal mill and the inlet temperature of the coal mill is as follows: Figure 5 and Figure 6 (Each data segment has two hours - 7200 data points) as shown.

[0137] The first step is to pre-process the variables in the figure, and use the bottom-up piecewise linear representation method to segment the 10 sample data into PLR segments. According to the trend change relationship and the actual mechanism relationship between the variables, the final judgment result is to select the five related variables of coal feed rate, mill inlet and outlet differential pressure, mill outlet temperature, exhaust motor current, and mill motor current to establish the feasible working domain boundary model of the coal mill pulverizing system.

[0138] Step 2: First, obtain the historical normal data of each associated variable. After processing and eliminating the unreasonable data parameters that deviate from the set value of the coal feeding rate, the four monitoring variables of the exhaust motor current x1, the mill motor current x2, the mill outlet temperature x3, the mill inlet and outlet differential pressure x4 and a certain coal feeding rate x5 (average value 29.13t / h) can be used to establish the feasible working range boundary model. Figure 7 Then the Z-score standardization method was used to standardize the original data into normalized data with a mean of 0 and a variance of 1.

[0139] Formula (4) and formula (5) are used to calculate the radius r of the hypersphere that can enclose all normal data points. The step angle of the search cone is set to α, and then the various indicators and the objective function F(α) in formula (6) are calculated as follows: Figure 8 As shown. The optimal deflection angle (step angle) of the search cone is 6.8°, the model fitness index is 0.9092, the comprehensive standard deviation index is 2.9637, and the radius difference index is 41.2553. Then let each angular coordinate φ j ,j=1,2,…,n-1 is equal to 0 and α, and all the permutations and combinations of angular coordinates are substituted into formula (18) to obtain the coordinates of the vertex of the infinitesimal element, and together with the origin O, they form the vertex set V of the n-dimensional space search cone. The search cone expression is obtained by substituting the data in the vertex set V into formulas (20), (21) and (22) to solve. Then, by substituting the n-1th angular coordinate φ n-1 The feasible working domain boundary model M is obtained by traversing the search with step α. The feasible working domain boundary model finally obtained is as follows: Figure 9 shown.

[0140] Step 3: First, according to formula (3), the original monitoring data points The preprocessing is to standardize the data point X(t0). According to formula (35), it is divided into normal data points and abnormal data points. For the abnormal data point, the simultaneous equations (26), (28), (32) and (33) are solved to obtain the normal data point X corresponding to the abnormal data point. M Then calculate x according to equations (26) and (28) j (t0) Line Y corresponding to the dynamic alarm threshold j , and then the straight line Y j The intersection of the two is obtained by solving the feasible working area boundary model simultaneously, and then substituting them into equations (30) and (31) to obtain the dynamic alarm threshold of the original data. as well as Let j = 1, 2, ..., n respectively, and repeat the above steps to solve all process variables x at the data point X(t0) j The dynamic alarm threshold value obtained finally Figure 10 shown.

[0141] Figure 10 The curve in the middle of each variable represents the standardized monitored operating value of the process variable. The upper and lower curves correspond to its high and low alarm thresholds, respectively. The bold curve on the right represents an abnormal operating state. This shows that the monitored data points initially operate within the feasible operating range, then gradually exceed their normal operating range at t = 4226 seconds. This indicates that the alarm system should issue an alarm signal at t = 4226 seconds.

[0142] The present invention also provides the following product embodiments:

[0143] An industrial multivariable alarm system based on search cone feasible working domain modeling, comprising:

[0144] a monitoring variable determination module configured to determine the associated variable determined based on the trend change as a monitoring variable;

[0145] The data point judgment module is configured to pre-process the data points of the acquired monitoring variables and search whether a search cone can contain the pre-processed data points in a feasible working domain boundary model pre-built based on historical monitoring variable data. If so, the search cone is a normal data point, otherwise it is an abnormal data point;

[0146] A dynamic alarm threshold determination module is configured to determine normal data points corresponding to abnormal data points, calculate a straight line corresponding to the dynamic alarm threshold based on the normal data points, calculate the intersection of the straight line and the feasible working area boundary model, and then determine the dynamic alarm threshold;

[0147] The alarm module is configured to trigger an alarm if the preprocessed data point is outside the feasible working domain boundary model, or if a certain variable exceeds the operating range represented by its corresponding alarm threshold, otherwise no alarm is triggered.

[0148] It can be seen from the embodiments that the present invention simplifies the calculation process and improves the practicability of the model by calculating abnormal data points by judging whether the abnormal data points are within the search cone model.

[0149] At the same time, the present invention adopts a method of preprocessing data to achieve data screening, fully utilizing the direct connection between the data of each variable, and only requires the historical normal data points of the process variables to form a closed area of ​​the convex hull model in the high-dimensional space, overcoming the shortcomings of the hyperellipsoid feasible working domain boundary model that the process variables are limited to satisfy the Gaussian distribution and the convex hull model that the historical normal data points of the process variables must form a convex closed area in the high-dimensional space, thereby increasing versatility.

[0150] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0151] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0152] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0155] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. An industrial multivariable alarm method based on search cone feasible working domain modeling, characterized by: The following steps are involved: Determine the associated variables determined based on trend changes as monitoring variables; The data points of the acquired monitoring variables are pre-processed, and a search cone is found in a feasible working domain boundary model pre-built based on the historical monitoring variable data to see whether there is a search cone that can contain the pre-processed data points. If so, it is a normal data point, otherwise it is an abnormal data point; Determine the normal data points corresponding to the abnormal data points; Based on the normal data points, calculate the straight line corresponding to the dynamic alarm threshold, calculate the intersection of the straight line and the feasible working area boundary model, and then determine the dynamic alarm threshold; If the preprocessed data point is outside the feasible working domain boundary model, or if a variable exceeds the operating range represented by its corresponding alarm threshold, an alarm is triggered; The specific process of pre-building a feasible work domain boundary model based on historical monitoring variable data includes: Preprocess the normal data points of the monitoring variables in the historical monitoring data, regard the feasible working domain composed of the monitoring variables as consisting of multiple search cones, and determine the side edge length of the search cone based on the preprocessed data; Set the optimal step angle of the search cone, select the values ​​of the corner coordinates of each search cone, and obtain the vertex coordinates of the infinitesimal element according to the arrangement and combination of all corner coordinates, and form a vertex set together with the origin; According to the vertex set, the search cone expression is determined, and the feasible working domain boundary model is obtained by traversal search; The search cone is represented by a convex hull in a high-dimensional space, and the vertex set of the convex hull is composed of the origin and the vertices of the hypercube at the bottom of the search cone; Find the data points that are in the current search cone based on all historical normal data points , and Middle distance origin The farthest data point is defined as the boundary data point , then the feasible working domain of the current search cone can be expressed as follows, and the expression that can describe the feasible working domain of the current part is temporarily stored in the set middle; The first Angular coordinates Stepping , when two adjacent angle coordinates The deviation angle is less than its upper limit (or ), and the angular coordinates The deflection angle is less than its upper limit (or ) and continue to calculate; otherwise, let 0 and , the offset angle The value step , and in End the traversal process when the constraints are not met; According to the above calculation sequence, the feasible working area boundary model is obtained. ; The specific process of setting the optimal search cone step angle includes: Define the vertex angle of two adjacent edges of the search cone as the step angle of the search cone , The value of is determined by the following objective function Determine when taking the minimum value: in, The fitness index of the model representing the boundary of the feasible working domain; Indicates the comprehensive standard deviation of the alarm thresholds of adjacent monitoring data points; Indicates the radius difference index of adjacent search cones; To facilitate calculation, we can take Middle distance origin The rounded value of the farthest data point is ; In formula (4), represents a normalized normal data point, Indicates rounding the operand upwards and returning the smallest integer greater than the operand. Indicates the modulus value of a vector representing a process variable. The calculation formula of the dimensional vector is: In the above formula Indicates the comprehensive standard deviation index of the high alarm threshold of adjacent monitoring data points, The comprehensive standard deviation index of the low alarm threshold of adjacent monitoring data points can be calculated by the following formula: In the above formula Indicates the number of data points for testing the feasible working domain boundary model, Respectively represent Test data points The amplitude fluctuation of the adjacent data points relative to the high alarm threshold and low alarm threshold of the reference data point can be calculated by the following formula: In the above formula Represent the test data points The high alarm threshold and low alarm threshold, Represent the test data points No. The high alarm threshold and low alarm threshold of adjacent data points; In the above formula, represents the number of search cones used to establish the boundary model of the feasible working domain, Indicates the The side length of the search cone.

2. The industrial multivariable alarm method based on search cone feasible working domain modeling according to claim 1, characterized in that: The specific process of determining the associated variables determined based on trend changes as monitoring variables includes: using a bottom-up piecewise linear representation method to perform piecewise linear representation on the sample data, obtaining the trend change relationship of each variable, and then determining the monitoring variables based on the trend change relationship and the actual relationship between each variable.

3. The industrial multivariable alarm method based on search cone feasible working domain modeling according to claim 1, characterized in that: The specific process of preprocessing includes: using vectors to represent the historical normal data points of each monitoring variable, and using the Z-score standardization method to standardize the data of each monitoring variable.

4. The industrial multivariable alarm method based on search cone feasible working domain modeling according to claim 1, characterized in that: Select the values ​​of the corner coordinates of each search cone respectively, and obtain the coordinates of the vertex of the infinitesimal element according to the arrangement and combination of all the corner coordinates. The specific process includes: setting each corner coordinate equal to 0 and the step angle , Search the angular coordinates of each edge of the cone in dimensional space, the angular coordinates Total The different permutations and combinations in the search for the infinitesimal element at the bottom of the cone correspond to each other. Vertices, according to the arrangement and combination of all angular coordinates, the coordinates of the infinitesimal vertex are obtained.

5. The industrial multivariable alarm method based on search cone feasible working domain modeling according to claim 1, characterized in that: According to the vertex set, the specific process of determining the search cone expression includes: finding the data points on all hyperplanes of the convex hull through the fast convex hull algorithm, and using the expression to express the first hyperplanes, index the on a hyperplane data points and substitute them into the expression, solve to obtain the normal vector and hyperplane offset, calculate all hyperplanes that form the convex hull one by one, obtain the complete expression of the hyperplane to represent the search cone, and obtain the search cone expression.

6. The industrial multivariable alarm method based on search cone feasible working domain modeling according to claim 1, characterized in that: Calculate the straight line corresponding to the dynamic alarm threshold, calculate the intersection of the straight line and the feasible working domain boundary model, and then determine the dynamic alarm threshold. The specific process includes: according to the straight line expression corresponding to the dynamic alarm threshold of a certain monitoring variable, the straight line expression is linked with the feasible working domain boundary model to obtain the intersection of the two, forming a set , according to the set The maximum and minimum values ​​determine the upper and lower limits of the dynamic alarm threshold.

7. The system of an industrial multivariable alarm method based on search cone feasible working domain modeling according to any one of claims 1 to 6, characterized in that: include: a monitoring variable determination module configured to determine the associated variable determined based on the trend change as a monitoring variable; The data point judgment module is configured to pre-process the data points of the acquired monitoring variables and search whether a search cone can contain the pre-processed data points in a feasible working domain boundary model pre-built based on historical monitoring variable data. If so, the search cone is a normal data point, otherwise it is an abnormal data point; A dynamic alarm threshold determination module is configured to determine normal data points corresponding to abnormal data points, calculate a straight line corresponding to the dynamic alarm threshold based on the normal data points, calculate the intersection of the straight line and the feasible working area boundary model, and then determine the dynamic alarm threshold; The alarm module is configured to trigger an alarm if the preprocessed data point is outside the feasible working domain boundary model, or if a certain variable exceeds the operating range represented by its corresponding alarm threshold, otherwise no alarm is triggered.

Images