A method for detecting abnormal data in sintering process based on information granulation

Abstract

The present invention discloses a method for detecting abnormal data in a sintering process based on information granulation. First, the present invention selects the time series data of the sintering endpoint of the sintering process as the detection object, and performs information granulation on the time series data according to the principle of reasonable granularity to obtain information granules. Then, a test statistic is constructed, the probability density function and the cumulative distribution function of the test statistic are calculated, and the test threshold is determined according to the confidence level. Finally, a certain section of the time series data of the sintering endpoint is tested. If the test statistic exceeds the test threshold, data anomalies exist in this section of the time series data. The present invention uses the information granulation method to provide an accurate and effective solution for the detection of abnormal data in the sintering process.

Description

Technical Field

[0001] The invention relates to a method for detecting abnormal data in a sintering process based on information granulation, and belongs to the field of abnormal data detection in industrial production processes. Background Art

[0002] Abnormal data is the external manifestation of abnormal working conditions in industrial production processes. Abnormal working conditions often lead to low output and poor quality of products. Effective detection of abnormal data can provide data support for abnormal working condition identification and provide guidance for parameter decision-making in industrial production processes. Abnormal data detection allows enterprise production managers and actual production operators to more intuitively grasp the operating status of industrial processes, discover abnormalities in production data in a timely manner, and provide a strong reference for optimizing and adjusting the production process.

[0003] As a traditional industrial process, the sintering process has always attracted the attention of scholars. It is a time-consuming process with complex procedures. Due to the large number of parameters, it is difficult for operators to adjust the operating status in real time. The detection of abnormal data in the sintering process mainly analyzes the time series data of the sintering endpoint, detects its characteristics that are different from normal data, and uses it as the standard for subsequent abnormal data detection. At present, the characteristics of the time series data of the sintering endpoint that are different from normal data are difficult to extract and describe, which brings great obstacles to abnormal data detection. Therefore, the development of abnormal data detection methods for sintering processes has important economic value.

[0004] Information granules are a collection of entities gathered together by similarity. Information granulation of time series is to aggregate time series data according to data compactness, and then construct information granules described by granule prototypes and boundaries. Information granulation of time series can discover meaningful and interpretable relationships and form an overall description of time series. In addition, dynamic characteristics of time series can be effectively extracted. Therefore, it is necessary to invent a method for detecting abnormal data in sintering process based on information granulation. Summary of the invention

[0005] The present invention discloses a method for detecting abnormal data of a sintering process based on information granulation. First, the present invention selects the time series data of the sintering endpoint of the sintering process as the detection object, and according to the principle of reasonable granularity, performs information granulation on the time series data to obtain information granules. Then, a test statistic is constructed, the probability density function and the cumulative distribution function of the test statistic are calculated, and the test threshold is determined according to the confidence level. Finally, a certain section of the time series data of the sintering endpoint is tested. If the test statistic exceeds the test threshold, then there is data anomaly in this section of the time series data. In order to achieve the above technical purpose, the present invention provides the following method.

[0006] A method for detecting abnormal data in a sintering process based on information granulation comprises the following steps:

[0007] Step 1: Select the time series data of the sintering end point of the sintering process as the detection object;

[0008] Step 2: According to the principle of reasonable granularity, the time series data is granulated to obtain information granules;

[0009] Step 3: Construct the test statistic and calculate the probability density function and cumulative distribution function of the test statistic;

[0010] Step 4: Determine the test threshold based on the confidence level;

[0011] Step 5: If the test statistic of a certain section of sintering endpoint time series data exceeds the test threshold, then this section of time series data has data anomalies.

[0012] Furthermore, the sintering endpoint selected in step 1 is P, and its time series data is expressed as P = {p 1 ,p 2 ,...,p N}, where N is the length of the time series.

[0013] Furthermore, step 2 specifically includes the following process:

[0014] (3-1) Obtain the differential time series. For the time series data P = {p 1 ,p 2 ,...,p N}, its first-order difference time series is ΔP={Δp 1 ,Δp 2 ,...,Δp N}, where Δp i =p i -p i-1 ,i=1,2,...,N(requires p 0 is known).

[0015] (3-2) Normalization of time series data. Suppose the normalized time series is Q = {q 1 ,q 2 ,...,q N} and the first-order difference time series is ΔQ={Δq 1 ,Δq 2 ,...,Δq N},in

[0016]

[0017] In the formula, p min and p max is the minimum and maximum value of the time series P, Δp min and Δp maxare the minimum and maximum values ​​of the time series ΔP.

[0018] (3-3) Obtain information granules according to the principle of reasonable granularity. Suppose the form of information granules is Ω = {(l q ,l Δq ),(h q ,h Δq )}, where h q and h Δq represents the upper bound of the information particle Ω in different directions, l y and l Δy Indicates the lower bound of information granule Ω in different directions. Below we define two functions to express the rationality and particularity of information granule Ω.

[0019]

[0020] In the formula, f 1 Used to measure the rationality of information particles, f 2 It is used to measure the specificity of information particles. card(·) represents the number of elements in the set, and e represents the natural logarithm. The following optimization goal is set:

[0021] max:f(l q ,l Δq ,h q ,h Δq )=f 1 ×f 2

[0022] In the formula, max:f(·) means to find the maximum value of f(·). q ,h Δq , l y and l Δy is the decision variable, and the above formula is the optimization target. Through the particle swarm optimization algorithm, we can get the information particle Ω={(l q ,l Δq ),(h q ,h Δq )}.

[0023] Furthermore, step three specifically includes the following process:

[0024] (4-1) Construct the test statistic. Let the test statistic t be expressed as,

[0025]

[0026] in, represents the average value of Q; N Ω =card{(q k ,Δq k )∣(q k ,Δqk )∈Ω,k=1,2,...,N}, which is the number of elements contained in the information granule.

[0027] (4-2) Calculate the probability density function of the test statistic. Use M (M is a positive integer and greater than 100) time series samples, perform step 2 and step (4-1) on them respectively, and obtain the test statistic T = {t 1 ,t 2 ,...,t M}. Using the kernel density estimation method, the probability density function of the test statistic t is

[0028]

[0029] Where w represents the window width; represents the estimated probability density function under w. In addition, the cumulative distribution function of the test statistic t is expressed as,

[0030]

[0031] Where u is the test threshold, and its value range is the same as the test statistic t.

[0032] Furthermore, in step 4, the confidence level is set to α (α∈[0.01,0.1]), and according to step 3, The u value equal to α / 2 and 1-α / 2 is t α / 2 and t 1-α / 2 , that is, the test threshold is t α / 2 and t 1-α / 2 .

[0033] Furthermore, in step 5, step 2 and step (4-1) are performed on a certain sintering endpoint time series data sample to obtain the test statistic t new If t new ≤t α / 2 or new ≥t 1-α / 2 , then there is data anomaly in this time series data.

[0034] The beneficial effects of the present invention based on its technical solution are:

[0035] (1) The present invention provides a method for detecting abnormal data in a sintering process based on information granulation, which uses a two-dimensional information granulation method to granulate time series information and extract characteristic information of the time series.

[0036] (2) The present invention provides a method for detecting abnormal data in a sintering process based on information granulation, which uses a kernel density estimation method to obtain a detection threshold for abnormal data. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is the probability density function of the sintering endpoint time series detection statistic.

[0038] Figure 2 is the cumulative distribution function and detection threshold of the sintering endpoint time series detection statistic.

[0039] Figure 3 is the time series data 1 of the sintering endpoint to be detected.

[0040] Figure 4 It is the time series data 2 of the sintering endpoint to be detected. DETAILED DESCRIPTION

[0041] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments:

[0042] Example:

[0043] A method for detecting abnormal data in a sintering process based on information granulation. First, the present invention selects the time series data of the sintering endpoint of the sintering process as the detection object, and granulates the time series data according to the principle of reasonable granularity to obtain information granules. Then, a test statistic is constructed, the probability density function and cumulative distribution function of the test statistic are calculated, and the test threshold is determined according to the confidence level. Finally, a certain section of the sintering endpoint time series data is tested. If the test statistic exceeds the test threshold, then this section of the time series data has data anomalies. The specific steps are as follows:

[0044] (1) Collect historical time series data of sintering production and sintering endpoint to obtain original sample data

[0045] The historical data of sintering production is stored in the local database of the industrial computer in the operation room in the form of daily reports. Based on the data in the daily reports, the historical time series data of the sintering end point for one month is collected with a sampling interval of 30 seconds to form the original sample data.

[0046] (2) Granulation of time series data

[0047] According to the steps described in patent claims 2 and 3, the length of each sintering endpoint time series is set to 10 minutes, that is, N = 20, and 5000 sintering endpoint time series data with a length of 20 are used to establish a sample database; and the time series is granulated to obtain the information granule Ω of each sample time series data = {(l q ,l Δq ),(h q ,h Δq )}.

[0048] (3) Constructing the test statistic

[0049] According to the steps described in claim 4, a test statistic is constructed. The probability density function of the test statistic is calculated to obtain Figure 1 The probability density function of the sintering endpoint time series detection statistic is shown in Figure 1. Then, the cumulative distribution function is calculated to obtain Figure 2 Cumulative distribution function of the sintering endpoint time series detection statistic is shown.

[0050] (4) Determine the test threshold

[0051] According to the steps described in claim 5, the confidence level is set to α = 0.05, and the following can be obtained: Figure 2 The test threshold t is shown 0.025 =3.9097 and t 0.975 =7.0193.

[0052] (5) Abnormal data detection experiment

[0053] According to the steps described in claim 6, two sintering endpoint time series data to be detected (such as Figure 3 and 4 As shown in Figure 2, an abnormal data detection experiment was conducted, in which the detection statistic t of the sintering endpoint time series data 1 to be detected new is 5.8932, because t 0.025 ≤t new ≤t 0.975 , so there is no data anomaly in the time series data 1 of the sintering endpoint to be detected. The detection statistic t of the time series data 2 of the sintering endpoint to be detected new is 3.6422, because t new ≤t 0.025 , so the data to be detected 2 has data anomalies.

[0054] The results of this embodiment show that the present invention can effectively detect abnormal data in the sintering process and has important economic and application value.

Claims

1. A method for detecting abnormal data in a sintering process based on information granulation, characterized in that The following steps are involved: Step 1: Select the time series data of the sintering end point of the sintering process as the detection object; Step 2: According to the principle of reasonable granularity, the time series data is granulated to obtain information granules; Step 3: Construct the test statistic and calculate the probability density function and cumulative distribution function of the test statistic; Step 4: Determine the test threshold based on the confidence level; Step 5: If the test statistic of a certain section of sintering endpoint time series data exceeds the test threshold, then this section of time series data has data anomalies; Step 2 specifically includes the following processes: (3-1) Obtain the differential time series. For the time series data P = {p1, p2, ..., p N }, its first-order difference time series is ΔP={Δp1,Δp2,...,Δp N }, where Δp i =p i -p i-1 ,i=1,2,...,N, requiring p0 to be known; (3-2) Normalization of time series data. Let the normalized time series be Q = {q1, q2, ..., q N } and the first-order difference time series is ΔQ={Δq1,Δq2,...,Δq N },in In the formula, p min and p max is the minimum and maximum value of the time series P, Δp min and Δp max are the minimum and maximum values ​​of the time series ΔP; (3-3) According to the principle of reasonable granularity, we obtain information granules. Suppose the form of information granules is Ω = {(l q ,l Δq ),(h q ,h Δq )}, where h q and h Δq represents the upper bound of the information particle Ω in different directions, l q and l Δq It represents the lower bound of information granule Ω in different directions. Two functions are defined below to represent the rationality and particularity of information granule Ω. In the formula, f1 is used to measure the rationality of information particles, f2 is used to measure the particularity of information particles, card(·) represents the number of elements in the set, and e represents the natural logarithm. The optimization goal of the following formula is set: max:f(l q ,l Δq ,h q ,h Δq )=f1×f2 In the formula, max:f(·) means to find the maximum value of f(·), with h q ,h Δq , l q and l Δq is the decision variable, and the above formula is the optimization target. Through the particle swarm optimization algorithm, we can get the information particle Ω={(l q ,l Δq ),(h q ,h Δq )}.

2. The method for detecting abnormal data in a sintering process based on information granulation according to claim 1, characterized in that: Step 3 specifically includes the following processes: (4-1) Construct a test statistic; let the test statistic t be expressed as, in, represents the average value of Q; N Ω =card{(q k ,Δq k )∣(q k ,Δq k )∈Ω,k=1,2,...,N} is the number of elements contained in the information granule; (4-2) Calculate the probability density function of the test statistic; use M (M is a positive integer and greater than 100) time series samples, perform step 2 and step (4-1) on them respectively, and obtain the test statistic T = {t1, t2, ..., t M }; Using the kernel density estimation method, the probability density function of the test statistic t is Where w represents the window width; represents the estimated probability density function under w; in addition, the cumulative distribution function of the test statistic t is expressed as, Where u is the test threshold, and its value range is the same as the test statistic t.

3. The method for detecting abnormal data in a sintering process based on information granulation according to claim 2, characterized in that: Step 4: Set the confidence level to α, where α∈[0.01,0.1]. According to step 3, find The value equal to α / 2 and 1-α / 2 is t α / 2 and t 1-α / 2 , that is, the test threshold is t α / 2 and t 1-α / 2 .

4. The method for detecting abnormal data in a sintering process based on information granulation according to claim 3 is characterized in that: Step 5: Perform step 2 and step (4-1) on a sample of sintering endpoint time series data to obtain the test statistic t new , if t new ≤t α / 2 or new ≥t 1-α / 2 , then there is data anomaly in this time series data.

Images