Fault detection method for industrial boiler based on dynamic weighted differential principal component analysis

Abstract

The application discloses an industrial boiler fault detection method based on dynamic weighted differential principal component analysis, which comprises the following steps: setting a fixed length time window to expand the boiler sample points to obtain an expanded boiler sample set; finding the first neighbor in space and the neighbor set of the first neighbor of the expanded boiler sample set, performing weighted differential processing to solve the multi-modal problem of the boiler sample; establishing a PCA model based on the processed boiler sample data, and collecting online data pairs of the boiler operation; judging whether the collected current data is abnormal based on the established PCA model, so that the adverse effects of dynamic characteristics and multi-modal characteristics on the boiler fault detection are eliminated; and performing fault detection on the boiler system by using a traditional PCA model and the dynamic weighted differential principal component analysis respectively, wherein experiments prove that the dynamic weighted differential principal component analysis method can effectively improve the fault detection rate, and has important significance in the production practice of the boiler system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fault diagnosis, and particularly relates to an industrial boiler fault detection method based on dynamic weighted differential principal component analysis. BACKGROUND

[0002] Industrial boilers are important energy conversion equipment, and with the development of modern industrial technology, their internal structures are increasingly integrated and complex. Once a fault occurs, it will cause huge energy waste, and even threaten life and property safety. Therefore, ensuring the safe and reliable operation of boilers has always been the focus of research in the control field. However, industrial boilers have time-varying characteristics, are highly complex, have many variables, and are strongly coupled, so it is difficult to establish an accurate mechanism model for fault diagnosis. Principal component analysis (PCA) method is a common numerical analysis theory, and is widely used in process monitoring and fault diagnosis. Since the PCA method only needs to use historical data of normal operation of the system to construct a model, without the need to establish an expert knowledge base and an accurate mathematical model, it can convert a group of variables that may be correlated into uncorrelated variables through orthogonal transformation, use as few comprehensive data as possible to contain as much change information as possible in the original data set, eliminate redundant information, and realize dimensionality reduction of the original data, which makes the PCA method widely used in fault diagnosis.

[0003] Although industrial boilers are a typical single-mode continuous control system, they exhibit obvious time-varying and multi-modal characteristics when the load changes. The traditional PCA method uses T 2 and SPE two statistical indicators for fault detection, but T 2 and SPE statistics use the premise that the data is single-mode structure, and the samples are independent of each other, so the fault detection effect is not ideal for systems with dynamic and multi-modal characteristics, which will cause a large number of false positives and false negatives. Therefore, it is an urgent problem to detect faults in industrial boilers with time-varying and multi-modal characteristics. SUMMARY

[0004] To solve the above problems, the technical scheme adopted by the application is as follows: an industrial boiler fault detection method based on dynamic weighted differential principal component analysis, comprising the following steps:

[0005] S1, setting a fixed length time window to expand the boiler sample points to obtain an expanded boiler sample set;

[0006] S2, finding the first neighbor in space and the neighbor set of the first neighbor of the expanded boiler sample set, and performing weighted differential processing to solve the multi-modal problem of the boiler sample;

[0007] S3, based on the processed boiler sample data to establish PCA model, while collecting boiler operation of online data pair;

[0008] S4, based on the established PCA model, to determine whether the collected current data exists abnormal.

[0009] Further: the setting fixed length of time window to extend the boiler sample point, the process of the extended boiler sample set is as follows:

[0010] To the sample data x t , set the length of N time window S(x t ) = {x1 (x) , x2 (x) ......x t (x) ......x N (x)} with it as the center, the sample data x t Extended into sample set S(x t ), describe the timing relationship between samples.

[0011] Further: the first neighbor and the first neighbor of the neighbor set in space of the extended boiler sample set, weighted difference processing, to solve the multi-modal problem of boiler sample process is as follows:

[0012] For each sample in the time window S(x t ), calculate the Euclidean distance between the sample and other sample points, find the first neighbor And the first neighbor of the first neighbor set

[0013] The first neighbor sample is processed by formula (1) to eliminate the influence of sample center drift on fault detection, and the new sample In the formula The mean of the first j nearest neighbor sample of the first neighbor, The weight is:

[0014]

[0015] Wherein: weight Determined by formula (2):

[0016]

[0017] In the formula: The Euclidean distance between the first neighbor And its jth nearest neighbor;

[0018] Weight The constraint condition of the formula is formula (3).

[0019]

[0020] Further, the process of establishing the PCA model based on the processed boiler sample data and collecting online data pairs of the boiler operation is as follows:

[0021] The data is processed by the time window expansion and weighted difference method to establish the PCA model and collect online data pairs of the boiler operation.

[0022] Further, the process of judging whether the collected current data is abnormal based on the established PCA model is as follows:

[0023] The online data collected is constructed into a time window, and the first neighbor and the neighbor set of the first neighbor of the sample in the time window are found in the training set, and the processed sample is obtained by using the weighted difference method.

[0024] Confirm whether the online data collected currently is abnormal, calculate the T 2 and SPE statistics of the processed online data of the boiler, and judge whether the T 2 and SPE statistics exceed the corresponding statistical threshold, if the T 2 and SPE statistics do not exceed the statistical threshold, the boiler is normal at this time, otherwise the boiler fails at this time.

[0025] An industrial boiler fault detection device based on dynamic weighted difference principal component analysis, comprising:

[0026] A data expansion module is configured to set a fixed length time window to expand the boiler sample points to obtain an expanded boiler sample set.

[0027] A processing module is configured to find the first neighbor and the neighbor set of the first neighbor of the expanded boiler sample set in space, and perform weighted difference processing to solve the multi-modal problem of the boiler sample.

[0028] An establishing module is configured to establish a PCA model based on the processed boiler sample data and collect online data pairs of the boiler operation.

[0029] The judging module is used for judging whether the collected current data is abnormal based on the established PCA model.The industrial boiler fault detection method based on the dynamic weighted differential principal component analysis has the following advantages: the dynamic characteristics and the multi-modal characteristics have no adverse effects on the boiler fault detection by the method;the traditional PCA model and the dynamic weighted differential principal component analysis are respectively used for the fault detection of the boiler system, and the experiments prove that the dynamic weighted differential principal component analysis method can effectively improve the fault detection rate, and has important significance in the production practice of the boiler system. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative labor under the premise of the drawings.

[0031] Figure 1 The flow chart of the fault diagnosis of the dynamic weighted differential principal component analysis method of the present application;

[0032] Figure 2 (a) is a scatter plot of original data, and (b) is a scatter plot of data processed by the dynamic weighted differential method;

[0033] Figure 3 (a) is a Gaussian distribution diagram of original data, and (b) is a Gaussian distribution diagram of data processed by the dynamic weighted differential method;

[0034] Figure 4 The schematic diagram of the boiler fault detection result corresponding to the traditional PCA method;

[0035] Figure 5 The schematic diagram of the boiler fault detection result corresponding to the dynamic weighted differential principal component analysis method. DETAILED DESCRIPTION

[0036] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict, and the present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0037] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The following description of at least one example embodiment is merely illustrative in nature and is in no way limiting on the application or its uses. Based upon a review of the embodiments in the present application, all other embodiments that would be obvious to one of ordinary skill in the art are intended to be within the scope of the present application.

[0038] It should be noted that the terminology used herein is for the purpose of describing specific embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0039] The relative arrangement of parts and steps, numerical expressions, and numerical values set forth in the examples are not intended to limit the scope of the application, unless otherwise specifically stated. It should be apparent that the dimensions of the various parts shown in the drawings are not to scale and are only meant to illustrate the general principles of the application. Techniques, methods, and apparatus known to those of ordinary skill are not discussed in detail, but are to be considered part of the disclosure, where appropriate. In all examples shown and discussed herein, any specific value should be interpreted as illustrative only and not as a limitation. Other examples of the example embodiments can have different values. It should be noted that like reference numerals and letters refer to like items in the following drawings, and thus, once an item is defined in one drawing, it need not be discussed further in subsequent drawings.

[0040] In the description of the present application, it should be understood that the orientation words such as "front, back, up, down, left, right", "transverse, vertical, perpendicular, horizontal" and "top, bottom" and the like indicated orientation or position relationship are generally based on the orientation or position relationship shown in the drawings, only for the convenience of describing the present application and simplifying the description, without the opposite indication, these orientation words do not indicate and imply that the indicated device or element must have a specific orientation or be constructed and operated in a specific orientation, therefore, it cannot be understood as a limitation on the scope of protection of the present application: the orientation words "inner, outer" refer to the inner and outer relative to the contour of each component.

[0041] For the purposes of the description, reference can be made to spatially relative terms, such as "above", "below", "upper", "lower", and the like, to describe the spatial relationship of one device or feature to another device or feature as depicted in the figures. It is to be understood that the spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientations depicted in the figures. For example, if a device depicted in the figures is inverted, then a device or element described as "above" or "up" another device or element would now be oriented "below" or "down" the other device or element. Thus, the exemplary term "above" can encompass both an orientation of above and below. The devices can be otherwise oriented (rotated 90 degrees or at other orientations) and the spatially relative descriptors used herein interpreted accordingly. The terms "first", "second", "third", etc. do not necessarily indicate any ordinal, chronological or other sequence unless expressly stated to do so.

[0042] In addition, it should be noted that the use of "first", "second", and the like words of distinction do not have a special meaning and are merely used to distinguish between corresponding parts, and therefore cannot be understood as limiting the scope of protection of the present application.

[0043] Figure 1 Flow chart of fault diagnosis of the dynamic weighted differential principal component analysis method of the present application;

[0044] The method of the present application specifically comprises:

[0045] S1, obtaining data of the boiler under normal working conditions as a training set; for each sample in the training set, a time window with a length of N is constructed around the sample to expand the boiler sample points, and the size of N is determined by a cross-validation method; if N is too large, it will cause calculation redundancy, and if N is too small, it will not be able to describe the time sequence relationship between the boiler sample points;

[0046] S2, finding the first neighbor in space and the neighbor set of the first neighbor of the expanded boiler sample set, performing weighted differential processing, and solving the multi-modal problem of the boiler sample;

[0047] Specifically, for each sample in the training set time window, find the first neighbor in space and the first neighbor of the first neighbor The size of K is determined by a cross-validation method, and then a weighted differential method is used for processing to eliminate the adverse effects of sample center drift on fault detection, to obtain new training set samples

[0048] S3, establishing a PCA model based on the processed boiler sample data, and simultaneously collecting online data pairs of the boiler operation;

[0049] S4, judging whether the collected current data is abnormal based on the established PCA model.

[0050] The steps S1 / S2 / S3 / S4 are sequentially executed.

[0051] The process of expanding the boiler sample points by setting a fixed-length time window is as follows:

[0052] For the collected sample data x t , a time window with a length of N is set therearound, the sample data is expanded into a sample set, and the time sequence relationship between the samples is described.

[0053] The process of finding the first neighbor in space and the neighbor set of the first neighbor of the expanded boiler sample set, and performing weighted difference processing to solve the multi-modal problem of the boiler sample is as follows:

[0054] For each sample in the time window S(x t ), find its first neighbor in space and the first neighbor of the first neighbor

[0055] The first neighbor sample is processed by weighted difference processing by using formula (1) to eliminate the influence of sample center drift on fault detection, and a new sample is obtained. In the formula, is the mean value of the first j nearest neighbor samples of the first neighbor, is the weight value:

[0056]

[0057] wherein the weight value is determined by formula (2):

[0058]

[0059] In the formula: is the Euclidean distance between the first neighbor and the jth nearest neighbor thereof;

[0060] The weight value The constraint condition of the formula is formula (3).

[0061]

[0062] The S3 includes the following steps:

[0063] S31, constructing a sample matrix X=(X ij )n ×m, where m is the number of sample variables, n is the number of samples, and is the standardization processing of eliminating the dimensional and order of magnitude restrictions on data.

[0064] S32, calculate the covariance matrix S of the same:

[0065]

[0066] S33, determine the number of principal components k by using the cumulative percent variance (CPV):

[0067]

[0068] Where: λ is the eigenvalue of S;

[0069] S34, establish a PCA model:

[0070] X i = X p +E = T i P i T +E (6)

[0071] T i = X i P i (7)

[0072] Where: X P is the projection of the sample vector X on the principal component subspace; T is the score matrix; P is the loading matrix; E is the residual matrix.

[0073] The S4 includes the following steps:

[0074] S41, construct a time window for the collected online data, find the first neighbor and the neighbor set of the first neighbor of the sample in the time window in the training set, and obtain the processed sample using a weighted difference method.

[0075] S42, confirm whether the currently collected online data is abnormal, calculate the T 2 and SPE statistics of the processed boiler online data, and determine whether the T 2 and SPE statistics exceed the corresponding statistical threshold, if the T 2 and SPE statistics do not exceed the statistical threshold, the boiler is normal at this time, otherwise the boiler fails at this time.

[0076] Specifically, the T 2 and SPE two statistical indicators are used to detect the boiler fault, and the principle involved is:

[0077] T 2The statistical quantity represents the amplitude and variation trend of the sampling point and the variation of the deviation from the principal component subspace, and the calculation formula is:

[0078]

[0079] Wherein, t i is the i-th row of the T matrix; λ is a diagonal matrix composed of the characteristic values corresponding to the first k principal component vectors.

[0080] T 2 The statistical quantity threshold calculation formula is:

[0081]

[0082] Wherein, a is the significance level, 1-a is the confidence level, F a (a, n-a) is the F distribution with the first degree of freedom a and the second degree of freedom n-a. If T i 2 ≤CL T2 , it indicates that the T 2 statistical quantity is normal, otherwise, it is a fault.

[0083] The SPE statistical quantity represents the deviation variation of the sampling point in the residual subspace, and the calculation formula is:

[0084] SPE i = X T (I-PP T ) T (I-PP T )X (10)

[0085] The SPE statistical quantity threshold calculation formula is:

[0086]

[0087]

[0088]

[0089] Wherein, C a is the confidence limit of the standard normal distribution with the confidence level a. If SPE i ≤CL SPE , it indicates that the SPE statistical quantity is normal, otherwise, it is a fault.

[0090] Subsequently, it is judged whether the T 2 and SPE statistical quantities of the online data exceed the corresponding statistical quantity thresholds, and if both do not exceed, it indicates that the boiler is running normally, otherwise, the boiler is in fault.

[0091] A kind of industrial boiler fault detection device based on dynamic weighted differential principal component analysis, comprising:

[0092] Data expansion module: for setting the time window of fixed length to the boiler sample point is expanded, and the extended boiler sample set is obtained;

[0093] Processing module: for finding the first neighbor and the neighbor set of first neighbor in space to the extended boiler sample set, and carrying out weighted differential processing, solve the multimodal problem of boiler sample;

[0094] Establishment module: for establishing PCA model based on the boiler sample data after processing, while collecting online data of boiler operation pair;

[0095] Judgment module: for judging whether the current data collected exists abnormity based on the PCA model established.

[0096] The embodiment of the present application is applied to a certain industrial boiler, and the fault diagnosis is carried out on the boiler system, and the main process parameters are shown in Table 1.

[0097] Table 1 boiler process parameter table

[0098] No. Variable Unit No. Variable Unit 1 Outlet flue gas temperature ℃ 7 Outlet flow rate t·h -1 ]]> 2 Outlet flue gas temperature ℃ 8 Induced draft fan speed r·s -1 ]]> 3 Outlet flue gas temperature ℃ 9 Induced draft fan speed r·s -1 ]]> 4 Outlet water temperature ℃ 10 Grate speed r·s -1 ]]> 5 Inlet water temperature ℃ 11 Coal feed speed r·s -1 ]]> 6 Furnace pressure MPa 12 Flue gas oxygen content ％

[0099] The embodiment of the present application collects 1500 groups of data under normal working condition as training set sample, collects 225 groups of online data as test set sample, and injects fault in 206-225 groups of data.

[0100] As shown in Figure 2 and Figure 3 , any two variable directions are drawn respectively to draw the original data and the data scatter diagram and Gaussian distribution diagram after dynamic weighted differential processing, Figure 2 (a) is the original data scatter diagram, (b) is the data scatter diagram after dynamic weighted differential processing;

[0101] Figure 3 (a) is the original data Gaussian distribution diagram, (b) is the data Gaussian distribution diagram after dynamic weighted differential processing;

[0102] Therefore, the data after dynamic weighted differential processing is single mode and obeys Gaussian distribution, which meets the precondition of using PCA method for fault detection.

[0103] The fault detection result using traditional PCA method is shown in Figure 4 , and the fault cannot be detected. The fault detection result of dynamic weighted differential principal component analysis is shown in Figure 5 , and all faults can be completely detected.

[0104] In summary, the application is directed to industrial boiler systems with dynamic characteristics and multi-modal characteristics, and a dynamic weighted differential principal component analysis method is proposed. Through the method, the adverse effects of the multi-modal characteristics and dynamic characteristics of the dynamic system on boiler fault detection are eliminated. The traditional PCA method and the dynamic weighted differential principal component analysis are used for fault detection on the boiler system respectively, and experiments prove that the dynamic weighted differential principal component analysis method can effectively improve the fault detection rate, and has important significance in the production practice of the boiler system.

[0105] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for industrial boiler fault detection based on dynamic weighted differential principal component analysis, characterized by, The method comprises the following steps: S1, setting a fixed length time window to extend the boiler sample points to obtain an extended boiler sample set; S2, finding the first neighbor and the neighbor set of the first neighbor in space for the extended boiler sample set, and performing weighted difference processing to solve the multi-modal problem of the boiler sample; S3, establishing a PCA model based on the processed boiler sample data, and collecting online data pairs of the boiler operation; S4, judging whether the collected current data is abnormal based on the established PCA model.

2. The industrial boiler fault detection method based on dynamic weighted differential principal component analysis according to claim 1, characterized in that: The process of setting a fixed length time window to extend the boiler sample points to obtain an extended boiler sample set is as follows: For the collected sample data x t A time window S(x) of length N is set with x as the center. t )={x1 (x) x2 (x) ......x t (x) ......x N (x) }, take the sample data x t Expanded into a sample set S(x) t ), which describes the temporal relationship between samples.

3. The method for industrial boiler fault detection based on dynamic weighted differential principal component analysis according to claim 2, characterized in that: The process of finding the first neighbor and the neighbor set of the first neighbor in space for the extended boiler sample set, and performing weighted difference processing to solve the multi-modal problem of the boiler sample is as follows: For each sample in the time window S(x t ), find its first neighbor in space and the first neighbor's set of K nearest neighbors The first neighbor samples are weighted and differentially processed by using formula (1) to eliminate the influence of sample center drift on fault detection, and new samples are obtained In the formula is the mean value of the first j nearest neighbor samples, is the weight wherein: weight is determined from equation (2): wherein: is the first nearest neighbor Euclidean distance between the point and its jth nearest neighbor Weights The constraint condition of the formula is formula (3).

4. The method for industrial boiler fault detection based on dynamic weighted differential principal component analysis according to claim 1, characterized in that: The process of establishing a PCA model based on the processed boiler sample data, and collecting online data pairs of the boiler operation is as follows: After the data is processed by the time window extension and weighted difference method, a PCA model is established, and online data pairs of the boiler operation are collected.

5. The method for industrial boiler fault detection based on dynamic weighted differential principal component analysis according to claim 1, characterized in that: The process of judging whether the collected current data is abnormal based on the established PCA model is as follows: S51, constructing a time window for the collected online data, finding the first neighbor and the neighbor set of the first neighbor in space for the samples in the time window in the training set, and obtaining the processed samples by using the weighted difference method. S52, confirming whether the current collected online data is abnormal, calculating T 2 and SPE statistics, and determining whether T 2 and SPE statistics exceed the corresponding statistical threshold, if T 2 and SPE statistics do not exceed the statistical threshold, the boiler is normal at this time, otherwise the boiler fails at this time.

6. An industrial boiler fault detection apparatus based on dynamic weighted differential principal component analysis, characterized by, It comprises: A data expansion module for setting a fixed length time window to extend the boiler sample points to obtain an extended boiler sample set; A processing module for finding the first neighbor and the neighbor set of the first neighbor in space for the extended boiler sample set, and performing weighted difference processing to solve the multi-modal problem of the boiler sample; An establishment module for establishing a PCA model based on the processed boiler sample data, and collecting online data pairs of the boiler operation; A judgment module for judging whether the collected current data is abnormal based on the established PCA model.

Images