A polar meteorological equipment fault detection method based on density peak and anomaly isolation

Abstract

The present invention provides a polar meteorological equipment fault detection method based on density peak and anomaly isolation. First, the noise data of the polar meteorological equipment electrical signal is removed by a density clustering-based method; secondly, the electrical signal working mode of the polar meteorological equipment during normal operation is mined out in the non-noise data through the density peak; finally, the real-time electrical signal data of the polar meteorological equipment and the electrical signal feature data during normal operation are used to jointly construct an isolation forest to perform real-time fault diagnosis on the polar meteorological equipment. At the same time, when constructing the isolation forest, isolated trees are selectively constructed through the segmentation ratio, thereby improving the efficiency of the algorithm to a certain extent, making it more able to meet the real-time requirements of digital twins. Finally, the test data of the precipitation meter under different low temperature environments were used to verify the effectiveness and certain advancement of the model.

Description

Technical Field

[0001] The present invention belongs to the field of digital twins, and in particular relates to a method for detecting faults of polar meteorological equipment. Background Art

[0002] At present, the polar regions have become an important strategic location for each country to conduct climate observation, energy mining, etc. At the same time, polar climate observation and data collection are also of great significance to polar scientific research. Therefore, we need to perform real-time fault detection on polar meteorological equipment. However, in the polar environment, due to the harsh environment and the prone to extreme low temperatures, it is very difficult to use manpower to monitor the operation and maintenance of polar meteorological equipment. Therefore, we need to apply digital twin technology to polar meteorological equipment and perform real-time fault diagnosis on polar meteorological equipment, so as to operate and maintain polar meteorological equipment as timely as possible, and to avoid the occurrence of more meteorological data missing due to long downtime caused by polar meteorological equipment failures.

[0003] However, most of the current digital twin fault detection models are constructed in the following two ways: based on equipment mechanism and based on data-driven. The equipment mechanism-based construction method requires detailed modeling of each device and detailed equipment parameters. At the same time, in polar environments, various environmental factors will also affect the equipment life degradation mechanism, which are difficult to characterize with mechanism models. However, in many methods that use data-driven methods to build digital twin fault detection models, the normal operation mode of the equipment is not first discovered, so the performance of its fault detection model in different environments is not the same. However, the polar environment is complex and changeable, and its meteorological equipment cannot always operate in the same environment. Therefore, it is not suitable to directly use the currently more common digital twin fault detection model construction method.

[0004] In summary, studying the digital twin model for fault detection of polar meteorological equipment has important research significance. Summary of the invention

[0005] Technical issues to be solved

[0006] In view of the problem that there are many equipment failures caused by low temperature in polar environments, the current common fault detection algorithms directly used on them cannot meet the fault detection needs well, such as LOF algorithm, iForest algorithm, etc. These methods all assume that the number of abnormal data accounts for a small number of total samples. The present invention provides a polar meteorological equipment fault detection method based on density peak and abnormal isolation, which can not only effectively detect the abnormality of the electrical signal of polar meteorological equipment, but also has high efficiency and can meet the real-time requirements of digital twins.

[0007] Technical Solution

[0008] A polar meteorological equipment fault detection method based on density peak and anomaly isolation, characterized by the following steps:

[0009] Step 1: Determine the temperature-related or non-temperature-related fault detection feature data of polar meteorological equipment.

[0010] Step 2: Use density-based clustering to remove noise points from the selected feature data for measuring whether polar meteorological equipment has failed.

[0011] Step 3: Obtain the electrical signal characteristic data of the polar meteorological equipment when it is working normally.

[0012] Step 4 uses the real-time electrical signal data of polar meteorological equipment and the electrical signal characteristic data during normal operation to selectively construct multiple effective isolated trees to form an isolated forest.

[0013] Step 5: Calculate the anomaly score of the real-time electrical signal data of polar meteorological equipment.

[0014] Step 6: Determine whether the polar meteorological equipment is faulty.

[0015] A further technical solution of the present invention: The method for determining temperature-related or non-temperature-related fault detection characteristic data of polar meteorological equipment in step 1 is:

[0016] Step 1.1 Calculate the correlation coefficient (Pearson correlation coefficient) between the electrical signal data of polar meteorological equipment and the ambient temperature.

[0017] In the formula, cov(X i ,T) is the electrical signal data of polar meteorological equipment X i Covariance with ambient temperature T.

[0018] σ(X i ) is the standard deviation of the electrical signal data of polar meteorological equipment, and σ(T) is the standard deviation of the ambient temperature.

[0019] Step 1.2 The electrical signal data X i As fault detection characteristic data related to temperature factors, otherwise as fault detection characteristic data related to occasional failure (non-temperature factors). When the temperature-related characteristic data has a short-term continuous abnormality, it is used as the basis for the decision to start the heating module of the polar meteorological equipment. When the non-temperature-related characteristic data has a short-term continuous abnormality, it is used as the basis for the decision to restart the polar meteorological equipment.

[0020] A further technical solution of the present invention: In step 2, the method for removing noise points in the selected characteristic data for measuring whether the polar meteorological equipment has failed is based on density clustering:

[0021] Step 2.1 The electrical signal data of polar meteorological equipment is X in a single dimension i =(x1,x2,…x m ) performs density-based clustering. First, specify the minimum sample size MinPts and the radius ε.

[0022] Step 2.2 Define core points, directly density-reachable points, and density-reachable points. If the number of electrical signals in the ε neighborhood of a certain electrical signal data is greater than or equal to MinPts, then the point is a core point, that is, |N ε (x i )|≥MinPts, where N ε (x i )={x i ∈D|Dist(x i ,x j )≤ε}, D is the set containing all electrical signal data, Dist(x i ,x j ) is the Euclidean distance between two points; if x j In x i In the ε neighborhood, at the same time x j is the core point, then x j and x i Direct density can be reached; for (x i ,x i+1 ,...x n ), if any two adjacent electrical signals are directly accessible, then x i and x n Density can reach.

[0023] Step 2.3 Select any data point x from the data set i , if for the parameters MinPts and ε, the selected data point x i As the core point, find all i The data points that can be reached by density form a cluster. i If it is a non-core point, another data point is selected and the above process is repeated until the data selection is completed. The data points that are not assigned to any cluster are noise points.

[0024] Step 2.4 According to the above density-based clustering method, the noise points of each dimension of data can be identified. After the noise points are identified, they are removed. That is, m data points are obtained after noise removal to obtain n noise points, and finally X is obtained. i =(x1,x2,...x n ), where n≤m.

[0025] A further technical solution of the present invention: The method for obtaining the electrical signal characteristic data of the polar meteorological equipment when it is working normally in step 3 is:

[0026] Step 3.1 Use the formula Calculate the density of each data point, n is the number of data points, d ij is the distance from data point i to j. χ is the indicator function, when x<0, χ(x)=1, otherwise χ(x)=0, d c is the cutoff distance. From this we can get ρ i =(ρ1,ρ2,...,ρ n ).

[0027] Step 3.2 Calculate the The point with the highest density can be taken From this we get δ i =(δ1,δ2,...δ n ).

[0028] Step 3.3 Set ρ i =(ρ1,ρ2,...,ρ n ) and δ i =(δ1,δ2,...δ n ) to get peek i =(ρ1δ1,ρ2δ2,...ρ n δ n )=(peek1,peek2,...peek n ). i Sort in descending order and then take peek i The first k data are used as the electrical signal characteristic data F when the polar meteorological equipment is working normally i =(F1,F2,…F k ), k<n.

[0029] A further technical solution of the present invention is as follows: In step 4, the method of selectively constructing multiple effective isolated trees using the real-time electrical signal data of polar meteorological equipment and the electrical signal characteristic data during normal operation to form an isolated forest is as follows:

[0030] Step 4.1 Add the electrical signal data to be detected to Fi =(F1,F2,…F k ) to obtain F iForest =(F1,F2,…F k ,F dec ). Select F iForest As training data, determine F iForest The maximum value F in the set max , minimum value F min .

[0031] Step 4.2 Select the cutting point P∈(F min ,F max ). Put the nodes with cutting points less than P on the left (forming the left set), and put the nodes with cutting points greater than or equal to P on the right (forming the right set). Let the number of left sets of the first split point be NUM left , the number of sets on the right is NUM right ,So If τ∈[1,10], the tree is considered to be an inferior tree, so it is skipped and the next tree is constructed recursively until 100 trees are constructed. The termination condition of tree growth is: each node of the tree has only one data or the maximum growth height of the tree is reached. The maximum height limit of the tree h is related to the number of subsamples related:

[0032] A further technical solution of the present invention: The method for calculating the abnormal score value of the real-time electrical signal data of the polar meteorological equipment in Step 5 is:

[0033] Step 5.1 Calculate the average path length of the constructed binary tree Where n is the number of samples when constructing the isolation tree, H(n-1)=ln(n-1)+0.5772156649.

[0034] Step 5.2 Set the electrical signal data to be detected F dec The average height of each tree in the isolated forest is E(h(x)). Then the electrical signal data to be detected F dec The anomaly score is Where x = F dec .

[0035] A further technical solution of the present invention: The method for determining whether the polar meteorological equipment fails in step 6 is:

[0036] Step 6.1 Determine the score value Score(F) for judging the polar meteorological equipment electrical signal abnormality. When S(x,n)>Score(F), it is judged that the polar meteorological equipment electrical signal is abnormal, that is, the equipment is faulty. When S(x,n)≤Score(F), it is judged that the polar meteorological equipment electrical signal is normal, that is, the equipment is in good condition.

[0037] Beneficial Effects

[0038] The present invention provides a polar meteorological equipment fault detection method based on density peak and anomaly isolation. First, the noise data of the polar meteorological equipment electrical signal is removed by a density-based clustering method; secondly, the electrical signal working mode of the polar meteorological equipment during normal operation is mined out in the non-noise data through the density peak; finally, the real-time electrical signal data of the polar meteorological equipment and the electrical signal feature data during normal operation are used to jointly construct an isolation forest to perform real-time fault diagnosis on the polar meteorological equipment. At the same time, when constructing the isolation forest, isolated trees are selectively constructed through the segmentation ratio, thereby improving the efficiency of the algorithm to a certain extent, making it more able to meet the real-time requirements of digital twins. Finally, the test data of the precipitation meter under different low temperature environments were used to verify the effectiveness and certain advancement of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0040] Figure 1 Correlation coefficient matrix of precipitation meter electrical signal data;

[0041] Figure 2 The operating voltage of the CMOS image sensor;

[0042] Figure 3 DBSCAN identifies the noise of the operating voltage;

[0043] Figure 4 CMOS voltage feature data extraction decision diagram;

[0044] Figure 5 The construction process of the fault detection method of polar meteorological equipment (precipitation meter);

[0045] Figure 6 A digital twin framework for fault detection of polar meteorological equipment (precipitation meter). DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0047] Embodiment 1:

[0048] Step 1: Determine the temperature-related or non-temperature-related fault detection feature data of the precipitation meter.

[0049] Step 1.1 calculates the correlation coefficients between the electrical signal data of the CMOS image sensor of the precipitation meter and the ambient temperature. The electrical signal data of the CMOS image sensor includes the operating voltage, scanning frequency, conversion gain, and full well capacity. Thus, a correlation coefficient matrix is ​​constructed, and the result is as follows: Figure 1 shown.

[0050] Step 1.2 According to Figure 1 As a result, the operating voltage can be used as non-temperature-related characteristic data, and the full well capacity can be used as temperature-related characteristic data. When the operating voltage is abnormal for a short period of time, it is believed that the equipment may have an occasional failure, and it is used as the basis for the decision to "restart the precipitation meter". When the full well capacity is abnormal for a short period of time, it is believed that the failure of the precipitation meter is caused by temperature factors, and it is used as the basis for the decision to "start the precipitation meter heating module".

[0051] Step 2: Here, select to perform fault detection on the working voltage of the CMOS image sensor of the precipitation meter.

[0052] Step 2.1 First, cluster the historical data of the operating voltage based on density, and then identify its noise points. The results before and after clustering are as follows: Figure 2 and Figure 3 As shown, Figure 3 In the figure, the lighter points are non-noise data, and the darker points are noise data.

[0053] Step 2.2 removes the noise data to obtain a non-noise voltage data set.

[0054] Step 3 obtains voltage characteristic data of the precipitation meter CMOS image sensor when it works normally.

[0055] Step 3.1 Calculate ρ for each data point in the non-noise voltage data. i =(ρ1,ρ2,...,ρ n ).

[0056] Step 3.2 Calculate the δ of each data point in the non-noise voltage data.i =(δ1,δ2,…δ n ).

[0057] Step 3.3: By ρ i =(ρ1,ρ2,...,ρ n ) and δ i =(δ1,δ2,...δ n ) constitutes a decision diagram, and the result is as follows Figure 4 As shown. i δ i The first 500 largest data are used as the voltage characteristic data F when the CMOS image sensor is working normally. i =(F1,F2,...F 500 ).

[0058] Step 4 uses the real-time voltage data of the CMOS image sensor and the voltage characteristic data of the CMOS image sensor when it is operating normally to selectively construct multiple effective isolated trees to form an isolated forest.

[0059] Step 4.1 Add the voltage data to be detected to F i =(F1,F2,...F 500 ) to obtain F iForest =(F1,F2,...F 500 ,F dec ). Select F iForest As training data, determine F iForest The maximum value F in the set max , minimum value F min .

[0060] Step 4.2 Select the cutting point P∈(F min ,F max ). Put the nodes with cutting points less than P on the left (forming the left set), and put the nodes with cutting points greater than or equal to P on the right (forming the right set). Let the number of left sets of the first split point be NUM left , the number of sets on the right is NUM right ,So If τ∈[1,10], the tree is considered to be an inferior tree, so it is skipped and the next tree is constructed recursively until 100 trees are constructed. The termination condition of tree growth is: each node of the tree has only one data or the maximum growth height of the tree is reached. The maximum height limit of the tree h is related to the number of subsamples related: The number of subsamples in this example

[0061] Step 5 calculates the abnormal score value of the real-time voltage data of the precipitation meter CMOS image sensor.

[0062] Step 5.1 Calculate the average path length of the constructed binary tree Where n is the number of samples when constructing an isolated tree, H(n-1)=ln(n-1)+0.5772156649. In this example, n=501.

[0063] Step 5.2 Set the voltage data F to be detected of the CMOS image sensor dec The average height of each tree in the isolated forest is E(h(x)). Then the voltage data to be detected F dec The anomaly score is Where x = F dec .

[0064] Step 6 determines whether the working voltage of the precipitation meter CMOS image sensor is abnormal, that is, whether the precipitation meter has equipment failure. Here, the score value for determining that the voltage signal is abnormal is 0.761.

[0065] Step 6.1 When S(x,n)>0.761, the voltage signal is judged to be abnormal, that is, the precipitation meter is faulty. When S(x,n)≤0.761, the voltage signal is judged to be normal, that is, the precipitation meter is in good condition.

[0066] The detection effect can be measured by the detection rate and false alarm rate. The detection rate is also called the true positive rate, which refers to the percentage of abnormal voltages detected by the fault detection algorithm of the precipitation meter to the actual number of abnormal voltages during the fault detection period of the precipitation meter. The false alarm rate is also called the false positive rate, which refers to the percentage of false voltage anomalies detected by the fault detection algorithm of the precipitation meter to the actual number of normal voltages during the fault detection period of the precipitation meter.

[0067] This embodiment selects 2211 normal voltage data and 147 abnormal voltage data of the precipitation meter CMOS image sensor. The fault detection algorithm of this patent is applied to it, and the effect is shown in Table 1.

[0068] Table 1 Detection effect of precipitation meter fault detection model

[0069]

[0070] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should be included in the protection scope of the present invention.

Claims

1. A polar meteorological equipment fault detection method based on density peak and anomaly isolation, characterized in that Here are the steps: S1: Determine temperature-related or non-temperature-related fault detection feature data for polar meteorological equipment; S1.1: Calculate the correlation coefficient between the electrical signal data of polar meteorological equipment and the ambient temperature In the formula, cov(X i ,T) is the electrical signal data of polar meteorological equipment X i Covariance with ambient temperature T; σ(X i ) is the standard deviation of the electrical signal data of polar meteorological equipment, σ(T) is the standard deviation of the ambient temperature; S1.2: The electrical signal data X i As fault detection characteristic data related to temperature factors, otherwise as fault detection characteristic data related to non-temperature factors; when the temperature-related characteristic data has a short-term continuous abnormality, it is used as the basis for the decision to start the polar meteorological equipment heating module; when the non-temperature-related characteristic data has a short-term continuous abnormality, it is used as the basis for the decision to restart the polar meteorological equipment; S2: Using density-based clustering to remove noise points from the selected feature data for measuring whether polar meteorological equipment has failed; S2.1: The electrical signal data of polar meteorological equipment is in a single dimension X i =(x1,x2,...x m ) performs density-based clustering; first specify the minimum sample size MinPts and radius ε; S2.2: Define core points, directly density-reachable points, and density-reachable points; if the number of electrical signals in the ε neighborhood of a certain electrical signal data is greater than or equal to MinPts, then the point is a core point, that is, |N ε (x i )|≥MinPts, where N ε (x i )={x i ∈D|Dist(x i ,x j )≤ε}, D is the set containing all electrical signal data, Dist(x i ,x j ) is the Euclidean distance between two points; if x j In x i In the ε neighborhood, at the same time x j is the core point, then x j and x i Direct density can be reached; for (x i ,x i+1 ,...x n ), if any two adjacent electrical signals are directly accessible, then x i and x n Density can reach; S2.3: Select any data point x from the data set i , if for the parameters MinPts and ε, the selected data point x i As the core point, find all i Density-reachable data points form a cluster; if the selected data object point x i If it is a non-core point, another data point is selected and the above process is repeated until the data selection is completed; the data point that is not assigned to any cluster is a noise point; S2.4: According to the above density-based clustering method, the noise points of each dimension of data can be identified. After the noise points are identified, they are removed. That is, after removing the noise from m data points, n noise points are obtained, and finally X is obtained. i =(x1,x2,...x n ), where n≤m; S3: Acquire the electrical signal characteristic data of polar meteorological equipment when it is working normally; S3.1: Use formula Calculate the density of each data point, n is the number of data points, d ij is the distance from data point i to j; χ is the indicator function, when x<0, χ(x)=1, otherwise χ(x)=0, d c is the cutoff distance; From this we can get ρ i =(ρ1,ρ2,...,ρ n ); S3.2: Calculate each data point The point with the highest density can be taken From this we get δ i =(δ1,δ2,...δ n ); S3.3: Set ρ i =(ρ1,ρ2,...,ρ n ) and δ i =(δ1,δ2,...δ n ) multiplied by peek i =(ρ1δ1,ρ2δ2,...ρ n δ n )=(peek1,peek2,...peek n ); peek i Sort in descending order and then take peek i The first k data are used as the electrical signal characteristic data F when the polar meteorological equipment is working normally i =(F1,F2,...F k ), k<n; S4: Use the real-time electrical signal data of polar meteorological equipment and the electrical signal characteristic data during normal operation to selectively construct multiple effective isolated trees to form an isolated forest; S4.1: Add the electrical signal data to be detected to F i =(F1,F2,...F k ) to obtain F iForest =(F1,F2,...F k ,F dec ), select F iForest As training data, determine F iForest The maximum value F in the set max , minimum value F min ; S4.2: Select the cutting point P∈(F min ,F max ): Put the nodes with cutting points less than P on the left to form the left set, and put the nodes with cutting points greater than or equal to P on the right to form the right set; let the number of left sets of the first split point be NUM left , the number of sets on the right is NUM right ,So If τ∈[1,10], the tree is considered to be an inferior tree, so it is skipped and the next tree is constructed recursively until 100 trees are constructed. The termination condition of tree growth is: each node of the tree has only one data or the maximum growth height of the tree is reached. The maximum height limit of the tree h is related to the number of subsamples. related: S5: Calculate the abnormal score value of the real-time electrical signal data of polar meteorological equipment; S5.1: Calculate the average path length of the constructed binary tree Where n is the number of samples when constructing the isolated tree, H(n-1)=ln(n-1)+0.5772156649; S5.2: Set the data of the electrical signal to be detected F dec The average height of each tree in the isolated forest is E(h(x)), so the electrical signal data to be detected is F dec The anomaly score is Where x = F dec ; S6: Determine whether polar meteorological equipment fails; S6.1: Determine the score value Score(F) for judging whether the electrical signal of the polar meteorological equipment is abnormal. When S(x,n)>Score(F), it is judged that the electrical signal of the polar meteorological equipment is abnormal, that is, the equipment is faulty; when S(x,n)≤Score(F), it is judged that the electrical signal of the polar meteorological equipment is normal, that is, the equipment is in good condition.

Images