A diagnostic method for dynamometer diagram
Through the improved clustering algorithm of multi-dimensional clustering performance evaluation and feature fusion of power graphs, the problems of poor clustering division effect and insufficient accuracy of power graph diagnosis in the existing technology are solved, and more efficient clustering effect is achieved and more accurate diagnosis.
Patent Information
- Application Number
- CN202411354129.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2024-12-10
- Estimated Expiration
- 2044-09-27
AI Technical Summary
In the prior art, clustering algorithms based on single evaluation parameters are difficult to achieve ideal clustering division on a diversity data set, and the accuracy and consistency of the diagnostic accuracy and consistency of the power diagram are difficult to ensure.
The multi-dimensional clustering performance evaluation method is used to evaluate the clustering results through five dimensions: compactness, overlap, cluster shape, similarity, and resolution, and the fuzzy C-mean value is improved by combining the time and time-frequency domain characteristics of the power diagram. Clustering algorithm to improve clustering effect and diagnostic accuracy.
It realizes more detailed clustering performance evaluation and more accurate power graph recognition results, which improves the effectiveness of clustering algorithm and the reliability of diagnosis.
Smart Images
Figure CN118861607B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of clustering algorithms, and in particular to a dynamometer diagram diagnosis method. Background Art
[0002] In the prior art, a variety of clustering effectiveness evaluation functions for the fuzzy C-means clustering algorithm have been proposed to evaluate and optimize the clustering effect. However, due to the diversity of data sets and structures, it is difficult for a effectiveness function based on a single evaluation parameter to achieve ideal clustering division on all data sets. Summary of the invention
[0003] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a dynamometer diagram diagnosis method.
[0004] In the first aspect, according to the multidimensional clustering performance evaluation method of the embodiment of the present application, the performance of the clustering algorithm is evaluated, the cluster center is obtained, and then the data points corresponding to the cluster center are obtained, and according to the data points, the following are obtained: the degree of membership corresponding to each data point; the maximum eigenvalue and the minimum eigenvalue of the clustering covariance matrix corresponding to each data point; so as to obtain the evaluation parameters of the clustering algorithm; the evaluation parameters include: compactness result, overlap result, cluster shape result, similarity result, separation result; a clustering validity function is established according to the evaluation parameters, the clustering validity function is optimized to obtain a first optimal result, and then the optimal number of clusters corresponding to the first optimal result is obtained, and the clustering algorithm before improvement is updated with the optimal number of clusters to obtain an improved clustering algorithm.
[0005] According to the multi-dimensional clustering evaluation method of the embodiment of the present application, the clustering results are evaluated through five dimensions: compactness, overlap, cluster shape, similarity, and separation, which is more comprehensive than the traditional single indicator evaluation method. Each indicator in the evaluation parameter reflects the performance of clustering from different angles, and can judge the quality of clustering more carefully to find the optimal number of clusters, and then improve the original clustering algorithm according to the evaluation results to obtain an improved clustering algorithm.
[0006] According to some embodiments of the present application, the clustering algorithm is a fuzzy C-means clustering algorithm.
[0007] Furthermore, the method specifically includes the following steps: S1a, initialization, setting the initial parameters of the fuzzy C-means clustering algorithm; S2a, obtaining the cluster centers, and then obtaining the data points corresponding to each cluster center, obtaining the membership corresponding to each data point according to the data points, and updating the cluster centers according to the membership; S3a, obtaining evaluation parameters according to the updated cluster centers, data points, memberships, and the maximum eigenvalue and minimum eigenvalue of the clustering covariance matrix corresponding to each data point; establishing a clustering validity function according to the evaluation parameters, optimizing the clustering validity function to obtain the first optimal result, and then obtaining the optimal number of clusters corresponding to the first optimal result; S4a, updating the fuzzy C-means clustering algorithm before improvement with the optimal number of clusters to obtain the improved fuzzy C-means clustering algorithm.
[0008] Further, step S3a includes the following steps: S3a1, according to the updated cluster center, data point, membership degree, and the maximum eigenvalue and minimum eigenvalue of the cluster covariance matrix corresponding to each data point, respectively obtain: the compactness result, overlap result, cluster shape result, similarity result, and separation result of the fuzzy C-means clustering algorithm to establish evaluation parameters; S3a2, according to the evaluation parameters obtained in S3a1, establish a clustering validity function; wherein, the sum of the compactness result, overlap result, and cluster shape result is the first result M, and the sum of the similarity result and the separation result is the second result H; obtain the clustering validity function F :
[0009] ;
[0010] The cluster validity function F Minimum value of F min As the first optimal result, the optimal number of clustering items corresponding to the first optimal result is output.
[0011] On the other hand, according to the dynamometer diagram diagnostic method of the embodiment of the present application, the following are respectively obtained: the time domain feature vector of the dynamometer diagram is a first vector; the time-frequency domain feature vector of the input signal corresponding to the dynamometer diagram is a second vector; the first vector and the second vector are input into the clustering algorithm improved by the multidimensional clustering performance evaluation method described in any of the aforementioned embodiments to obtain the operating condition classification result; wherein the dynamometer diagram is a load-displacement dynamometer diagram.
[0012] According to the dynamometer diagram diagnostic method of the embodiment of the present application, the time-frequency domain features of the input signal and the time-domain features of the dynamometer diagram are respectively obtained to better represent the corresponding working conditions. Therefore, the first vector corresponding to the time-domain features and the second vector corresponding to the time-frequency domain features are used as inputs of the improved clustering algorithm, which can obtain more accurate results, and the improved clustering algorithm can output more accurate classification results than the ordinary clustering algorithm, thereby obtaining a more reliable dynamometer diagram recognition result.
[0013] According to some embodiments of the present application, in the load-displacement dynamometer diagram: the suspension point load is the vertical coordinate and the bare rod displacement is the horizontal coordinate; the load-displacement dynamometer diagram is converted into a 2-D dynamometer diagram to obtain a first vector, specifically including: obtaining the time corresponding to each suspension point load in the load-displacement dynamometer diagram as the horizontal coordinate, and the suspension point load as the vertical coordinate; performing time domain analysis on the 2-D dynamometer diagram to obtain the first vector.
[0014] Furthermore, frequency domain analysis and time domain analysis are performed on the input signal to obtain the second vector, specifically including: using adaptive noise complete set empirical mode decomposition to process the load signal to obtain the second vector.
[0015] Furthermore, the first vector and the second vector are combined and input into the improved clustering algorithm, and the output operating condition classification results include: normal operating condition, gas-affected operating condition, insufficient fluid supply operating condition, reduction gearbox failure operating condition, and sand production operating condition.
[0016] Furthermore, the dynamometer diagram diagnostic method specifically includes the following steps: S100, collecting measured dynamometer diagram data; S200, normalizing the load-displacement dynamometer diagram and converting the load-displacement dynamometer diagram into a 2-D dynamometer diagram; S300, extracting the time domain features of the 2-D dynamometer diagram as a first vector; S400, obtaining the time-frequency domain features of the input signal as a second vector; S500, combining the first vector obtained in S300 and the second vector obtained in S400 to obtain an operating condition vector, and inputting the operating condition vector into an improved clustering algorithm to obtain an operating condition classification result.
[0017] Further, S300 specifically includes: S310, performing time domain analysis on the input signal, and extracting the time domain feature vector as:
[0018]
[0019] The time domain feature vector is taken as the first vector; where, is the average value of the input signal, is the quantile of the input signal, is the triple mean of the input signal, is the mean square value of the input signal, is the variance of the input signal, is the maximum amplitude of the input signal, is the minimum amplitude of the input signal, is the peak value difference of the input signal.
[0020] S320, Kurtosis characteristics of intrinsic mode function ; In the formula, ,in, is the kurtosis coefficient of the first three intrinsic mode function component signals; the intrinsic mode function is Hilbert transformed to extract the marginal spectrum frequency characteristics ; In the formula, is the centroid frequency in the marginal spectrum, is the mean square frequency in the marginal spectrum, is the frequency variance in the marginal spectrum; the two extracted characteristic parameters constitute the time-frequency domain feature vector as the second vector.
[0021] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:
[0023] Figure 1 is a flow chart of a multi-dimensional performance evaluation method according to an embodiment of the present application;
[0024] Figure 2 is a load-displacement dynamometer diagram according to an embodiment of the present application;
[0025] Figure 3 is a 2-D dynamometer diagram according to an embodiment of the present application;
[0026] Figure 4 It is a flow chart of the dynamometer diagram diagnosis method according to an embodiment of the present application. DETAILED DESCRIPTION
[0027] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.
[0028] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.
[0029] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions.
[0030] There are many clustering effectiveness evaluation functions for the fuzzy C-means clustering algorithm (FCM algorithm) in the prior art, which are used to evaluate and optimize the clustering effect. However, since the algorithm usually processes data sets with diverse types and structures, it is difficult to achieve ideal clustering division on all data sets.
[0031] In order to comprehensively evaluate the quality of clustering results, this application constructs a multi-dimensional clustering performance evaluation method to improve the clustering effect.
[0032] According to the multi-dimensional clustering performance evaluation method of the embodiment of the present application, the performance of the clustering algorithm is evaluated. First, the cluster center is obtained, and then each data point corresponding to the cluster center is obtained. According to the data points, the following is obtained:
[0033] The membership degree corresponding to each data point, the maximum eigenvalue and the minimum eigenvalue of the clustering covariance matrix corresponding to each data point are used to obtain the evaluation parameters of the clustering algorithm, and the evaluation parameters include: compactness result, overlap result, cluster shape result, similarity result, and separation result. A clustering validity function is established according to the evaluation parameters, and the clustering validity function is optimized to obtain the first optimal result, and then the optimal number of clusters corresponding to the first optimal result is obtained. The clustering algorithm before improvement is updated with the optimal number of clusters to obtain the improved clustering algorithm.
[0034] According to the multi-dimensional clustering performance evaluation method of the embodiment of the present application, the clustering results are evaluated through five dimensions: compactness, overlap, cluster shape, similarity, and separation, which is more comprehensive than the traditional single indicator evaluation method. Each indicator in the evaluation parameter reflects the performance of clustering from different angles, and can judge the quality of clustering more carefully to find the optimal number of clusters, and then improve the original clustering algorithm according to the evaluation results to obtain an improved clustering algorithm.
[0035] According to a further embodiment of the present application, in combination Figure 1 It is understood that the clustering algorithm is the fuzzy C-means clustering algorithm. Next, the fuzzy C-means clustering algorithm is described in detail by taking the multi-dimensional clustering performance evaluation method as an example. The method specifically includes the following steps:
[0036] S1a, initialization, setting the initial parameters of the fuzzy C-means clustering algorithm as the initial settings of the preliminary fuzzy C-means clustering algorithm.
[0037] S2a, obtain the cluster center of the clustering algorithm, then obtain the data points corresponding to each cluster center, obtain the current membership degree according to each data point, and update the cluster center according to the membership degree; in this step, the characteristics of the cluster can be more accurately represented by recalculating the center of each cluster.
[0038] Specifically, each data point is assigned a membership degree during the clustering process to indicate the degree of relationship between the membership degree and each cluster center. Updating the cluster centers based on the membership degree can optimize the clustering results and improve the accuracy of clustering.
[0039] S3a, obtaining the evaluation parameters according to the updated cluster centers, data points, membership degrees, and the maximum eigenvalue and minimum eigenvalue of the cluster covariance matrix corresponding to each data point; then establishing a cluster validity function according to the evaluation parameters, optimizing the cluster validity function to obtain a first optimal result, and then obtaining an optimal number of clusters corresponding to the first optimal result.
[0040] It is worth mentioning that since the fuzzy C-means clustering algorithm is used to divide data points into multiple fuzzy clusters and allows each data point to belong to multiple clusters, it allows sample points to have different memberships to multiple cluster centers, thereby enabling it to construct a multi-dimensional evaluation method to evaluate the structural characteristics of the data.
[0041] S4a, updating the fuzzy C-means clustering algorithm with the optimal number of clusters to obtain an improved fuzzy C-means clustering algorithm
[0042] More specifically, step S3a may include the following steps:
[0043] S3a1, according to the updated cluster center, data point, membership degree, and the maximum eigenvalue and minimum eigenvalue of the cluster covariance matrix corresponding to each data point, respectively obtain: the compactness result, overlap result, cluster shape result, similarity result, and separation result of the fuzzy C-means clustering algorithm to establish evaluation parameters.
[0044] S3a2, based on the five types of results obtained in S3a1, a clustering validity function is established. For better explanation, the sum of the compactness result, overlap result, and cluster shape result is defined as the first result, and the sum of the similarity result and separation result is defined as the second result. The clustering validity function is expressed as F , the first result is represented by M and the second result is represented by H.
[0045] but:
[0046] ;
[0047] The cluster validity function F Minimum value of F min As the first optimal result, the optimal number of clustering items corresponding to the first optimal result is output.
[0048] S4a, after taking the optimal number of clusters as the number of clustering items of the clustering algorithm, an improved fuzzy C-means clustering algorithm is established.
[0049] For a better understanding, the multi-dimensional clustering performance evaluation method is explained with a specific example.
[0050] Exemplary, combined Figure 1 To understand, the multi-dimensional clustering performance evaluation method includes the following steps:
[0051] S1a, initialization parameters, the parameters that need to be initialized include: maximum number of clusters , maximum number of iterations , fuzzy index m, membership matrix U, cluster center V, termination threshold , and the initial number of clusters c.
[0052] in, n is the total number of data points, and the value range of the fuzzy index m is: .
[0053] S2a, update the fuzzy partition matrix and cluster centers and get ;in, t Representative t , in the formula Used to judge without special meaning. Is it less than ,like , then proceed to S230. , then continue updating.
[0054] S3a. Construct a multi-dimensional clustering evaluation function for optimizing the fuzzy C-means clustering algorithm. The influencing parameters of the multi-dimensional clustering evaluation function include compactness, overlap, cluster shape, similarity, and separation. The multi-dimensional clustering evaluation function is optimized to obtain the first optimal result, and then the optimal number of clusters corresponding to the first optimal result is obtained.
[0055] Specifically, in S3a1:
[0056] Define the compactness result as c;
[0057]
[0058] Where: represents the cluster center The corresponding data points The weighted sum of squares of the distances between Represents data points Cluster Center The square of the degree of membership, represents the degree of membership, is the total number of data points.
[0059] In the subsequent formula, i,j,k are all arbitrary numbers, indicating the i, j , k indivual.
[0060] The smaller the compactness, the closer the data points in the cluster are to its cluster center, and the more compact the cluster is.
[0061] Define the resulting overlap as o:
[0062]
[0063] Where: and Respectively represent and The cluster centers are The membership degree of each data point is is the total number of data points.
[0064] A low value of this indicator indicates that the clusters in the clustering results have a high degree of differentiation, that is, the data points between different clusters have a large difference in membership, resulting in clearer boundaries between clusters. Conversely, a high value indicates that the data points have close membership between different clusters, resulting in blurred boundaries and high overlap between clusters. In this way, the formula provides an effective means to quantify the differentiation between clusters in the clustering results, which helps to evaluate the performance of fuzzy clustering algorithms when dealing with data sets with unclear boundaries.
[0065] Define the resulting cluster shape as cs:
[0066]
[0067] Where: is the maximum eigenvalue of the cluster covariance matrix, is the minimum eigenvalue of the cluster covariance matrix, i Indicates i indivual.
[0068] This formula evaluates the shape of the cluster by calculating the ratio of the maximum and minimum eigenvalues of each cluster covariance matrix. The smaller the ratio, the more spherical the cluster shape is and the more uniform the distribution of points inside the cluster is.
[0069] The similarity result is defined as sm:
[0070]
[0071] Where: Indicates Data points for The membership degree of the cluster centers is is the total number of data points.
[0072] The similarity module represents the selection of the cluster with the smallest sum of squared membership from all clusters c. This operation aims to find the cluster with the lowest similarity (i.e., the least similar internally) in the entire cluster set. The larger the value of , the higher the degree of membership of the data points within the cluster and the stronger the similarity.
[0073] Define the resolution result as se:
[0074]
[0075] and Corresponding to i and k Cluster centers.
[0076] Weighing different cluster centers and The larger the distance, the better the separation between cluster centers, and the more likely the clustering results will have better discrimination.
[0077] In this embodiment, the clustering validity function is:
[0078] F = ;
[0079] Specifically, the definition is as follows:
[0080] F ;
[0081] S3a2, specifically: , where c=(2,…,c max ), using the FCM algorithm to calculate the clustering validity function F Minimum value of F min And get the optimal solution of c. Specifically, if c <c max ,but:
[0082] After updating the membership matrix U, the cluster centers V, and the initial number of clusters c, S220 is performed again.
[0083] If c ≥ c max , then select (Right now F min ) corresponds to the optimal number of clusters , and the optimal number of clusters is used as the number of clusters of the improved clustering algorithm.
[0084] For a better understanding, the following is an exemplary description of how to obtain the compactness result, overlap result, cluster shape result, similarity result, and separation result.
[0085] It can be understood that the evaluation method of this embodiment determines the optimal number of clusters by optimizing the clustering validity function, and can dynamically determine the optimal number of clusters according to the structural characteristics of the data itself to obtain a more accurate classification result.
[0086] On the other hand, since manual analysis of the dynamometer diagram determines the downhole working conditions through experience and the characteristics of the dynamometer diagram, the accuracy and consistency of the diagnosis results are difficult to guarantee. To this end, technical personnel in the field use computer technology to classify or identify the dynamometer diagram. However, among the various methods of diagnosing the dynamometer diagram based on computer technology, since the traditional dynamometer diagram has the characteristics of simple images and few feature points, directly inputting the traditional dynamometer diagram image into the algorithm will, to a certain extent, lead to low feature utilization, large amount of data calculation, and low working condition recognition accuracy.
[0087] Therefore, an embodiment of the present application also discloses a dynamometer diagram diagnosis method, which performs diagnosis based on the multidimensional clustering performance evaluation method described in the aforementioned embodiment, converts the dynamometer diagram curve into the time domain, constructs a 2-D dynamometer diagram, extracts the time domain features and time-frequency domain features of the 2-D dynamometer diagram curve to improve feature utilization, and combines fuzzy membership and the geometric structure of the data set to achieve accurate and rapid classification of the dynamometer diagram.
[0088] Therefore, according to the dynamometer diagram diagnostic method of the embodiment of the present application, the time domain feature vector of the dynamometer diagram is obtained as the first vector; the time-frequency domain feature vector of the input signal corresponding to the dynamometer diagram is obtained as the second vector. Then the first vector and the second vector are input into the clustering algorithm improved by the multi-dimensional clustering performance evaluation method described in any of the aforementioned embodiments to obtain the working condition classification result.
[0089] According to the dynamometer diagram diagnostic method of the embodiment of the present application, the time-frequency domain features of the input signal and the time-domain features of the dynamometer diagram are respectively obtained to better represent the corresponding working conditions. Therefore, the first vector corresponding to the time-domain features and the second vector corresponding to the time-frequency domain features are used as inputs of the improved clustering algorithm, which can obtain more accurate results, and the improved clustering algorithm can output more accurate classification results than the ordinary clustering algorithm, thereby obtaining a more reliable dynamometer diagram recognition result.
[0090] Next, a detailed description is given of the case where the dynamometer diagram is a load-displacement dynamometer diagram, wherein in the load-displacement dynamometer diagram: the suspension point load is the ordinate and the bare rod displacement is the abscissa.
[0091] For example, the construction drawing is a load-displacement dynamometer diagram of a pumping well. The load-displacement dynamometer diagram is drawn as a dynamometer diagram image by using the bare rod displacement data of the pumping unit as the horizontal coordinate and the suspension point load as the vertical coordinate. Specifically, each data point in the image represents the real-time data collected at the production site under specific working conditions, and the numerical pairs of the bare rod displacement and the load form a one-to-one correspondence. These binary numerical pairs constitute the two-dimensional discrete point set required to draw the dynamometer diagram.
[0092] Further, the load-displacement dynamometer diagram is converted into a 2-D dynamometer diagram to obtain a first vector and the input signal is subjected to frequency domain analysis and time domain analysis to obtain a second vector, specifically including:
[0093] The time corresponding to each suspension point load in the load-displacement dynamometer diagram is obtained as the horizontal axis, and the suspension point load is used as the vertical axis; the 2-D dynamometer diagram is analyzed in the time domain to obtain the first vector; the load signal is processed using the adaptive noise complete set empirical mode decomposition to obtain the second vector.
[0094] It is worth mentioning that the diagnosis of the load-displacement dynamometer diagram corresponding to the pumping well requires the classification of the working conditions first. Since the fault types of the pumping well are diverse, the classification results of various working conditions of the pumping well are not only reflected in the time domain of the dynamometer diagram, but also in the time-frequency domain of the dynamometer diagram. Therefore, it is necessary to fuse and normalize multiple feature information to accurately and concisely judge the working condition type.
[0095] Therefore, we continue to use the load-displacement diagram of the pumping well as an example to illustrate the construction drawing. Figure 4 To understand, the specific steps include:
[0096] S100, collecting measured dynamometer diagram data.
[0097] Specifically, the displacement data of the polished rod of the pumping well obtained by using the SGT-2000 intelligent dynamometer installed at the hanging point of the polished rod of the pumping unit is used as the abscissa, and the hanging point load data at the hanging point of the polished rod of the pumping unit is obtained as the ordinate.
[0098] S200, normalizing the load-displacement dynamometer diagram, and converting the load-displacement dynamometer diagram into a 2-D dynamometer diagram.
[0099] Specifically, in the process of drawing the load-displacement dynamometer diagram, a maximum and minimum value normalization method is used for normalization processing.
[0100] More specifically, the drawing of the load-displacement dynamometer diagram can be achieved through the Matlab tool. Set the drawing parameters LineColor to black and BackgroundColor to white to generate a black and white binary image. This representation method can not only clearly show the relationship characteristics between load and displacement, but also effectively support subsequent image analysis and processing. In this way, the working status and load distribution of the sucker rod system can be intuitively observed, providing strong visual support for the dynamic performance evaluation and fault diagnosis of the pumping unit.
[0101] However, the load-displacement dynamometer diagram is simple and has few feature points. Directly inputting the image into the algorithm may, to a certain extent, lead to low feature utilization and large amount of data calculation.
[0102] Therefore, in this embodiment, the load-displacement dynamometer diagram is converted into a 2-D (2- Dimensional) dynamometer diagram. The conversion method is described in detail below:
[0103] Combination Figure 2-Figure 3 To understand, first, theoretical analysis of the load-displacement dynamometer diagram is performed to determine the four points A, B, C, and D. Figure 2 The circled area in .
[0104] Will Figure 2 The curve in is divided into the first segment and the second paragraph , first paragraph represent Figure 1 The second segment of the path from A to B and then to C represent Figure 1 The path from C to D and then back to A.
[0105] According to the sampling period of the load-displacement dynamometer diagram, the specific time point corresponding to each load value can be determined, and then the first segment Load on upper stroke and second section The load of the downstroke is sampled using the interpolation method. After sampling, the final value is Figure 3 The two-dimensional dynamometer diagram shown is a 2-D dynamometer diagram, in which the suspension point load is used as the ordinate and the corresponding time is used as the abscissa. Figure 2 A, B, C, D in are converted to Figure 3 The corresponding A, B, C, and D in the figure.
[0106] It can be understood that converting the dynamometer diagram into a 2-D dynamometer diagram and then extracting the time domain features and time-frequency domain features of the 2-D dynamometer diagram curve solves the shortcomings of the traditional dynamometer diagram image input algorithm, such as low feature utilization and large amount of data calculation.
[0107] S300: extracting the time domain feature of the 2-D dynamometer diagram as a first vector.
[0108] Specifically, the input signal is analyzed in the time domain to extract the following time domain features:
[0109] S311. Get the average value, and use the average value as the first parameter, defined as .
[0110] S312. Get the quantile, and use the quantile as the second parameter, defined as .
[0111] S313. Obtain three means, and use the three means as the third parameter, defined as , where the three means are the weighted means of the upper quartile, median, and lower quartile.
[0112] S314. Obtain the mean square value, and use the mean square value as the fourth parameter, defined as .
[0113] S315. Obtain the variance, and use the variance as the fifth parameter, defined as .
[0114] S316. Obtain the maximum amplitude, and use the maximum amplitude as the sixth parameter, defined as .
[0115] S317, obtain the minimum amplitude, and use the minimum amplitude as the seventh parameter, defined as .
[0116] S318, obtaining the peak difference, and taking the peak difference as the eighth parameter, defined as .
[0117] According to the first to eighth parameters, the time domain feature vector of the 2-D dynamometer diagram after time domain feature extraction is obtained:
[0118]
[0119] S400: Obtain time-frequency domain features of an input signal as a second vector.
[0120] In order to analyze the 2-D dynamometer diagram in time domain and accurately show the working status of the pumping unit under certain faults, the time-frequency domain features of the 2-D dynamometer diagram are extracted to improve the feature analysis of the 2-D dynamometer diagram.
[0121] Specifically, the load signal is processed by Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to obtain the kurtosis characteristics of the Intrinsic Mode Function (IMF). , where ,in, is the kurtosis coefficient of the component signals of the first three intrinsic mode functions (Intrinsic Mode Function).
[0122] It is worth noting that when performing the Hilbert transform, the kurtosis coefficients of the component signals of the first three intrinsic mode functions can be obtained through the empirical mode decomposition method. Similarly, the intrinsic mode function is Hilbert transformed to extract the marginal spectrum frequency characteristics. ; In the formula, is the centroid frequency in the marginal spectrum, is the mean square frequency in the marginal spectrum, is the frequency variance in the marginal spectrum.
[0123] Specific:
[0124] .
[0125] S500, combining the first vector obtained in S300 and the second vector obtained in S400 and inputting them into an improved clustering algorithm to obtain a working condition classification result.
[0126] Specifically, the time domain feature vector S obtained in S300 and the time-frequency domain feature vector X obtained in S400 are combined to obtain the corresponding operating condition vector Z=[S,X], and the operating condition vector Z=[S,X] is input into the improved fuzzy C-means clustering algorithm to obtain a dynamometer diagram operating condition recognition model based on multi-feature extraction and improved fuzzy C-means clustering algorithm, and then the operating condition classification result is output.
[0127] In this embodiment, the working condition classification results include:
[0128] Normal operating conditions, gas-affected operating conditions, insufficient fluid supply conditions, reduction gearbox failure conditions, and sand production conditions.
[0129] It is understandable that by analyzing the 2-D dynamometer diagram in the time domain and time-frequency domain, the feature extraction method is enriched, the accuracy of working condition identification is improved, and the evaluation criteria of key characteristic values are obtained by computer calculation, which removes the influence of subjective factors on the diagnosis results and ensures the identification accuracy. It also breaks through the technical bottleneck of the traditional dynamometer diagram working condition identification, has strong engineering practicality, low system development cost, and has good promotion and application value.
[0130] In addition, the obtained operating condition vector Z=[S,X] is input into the improved fuzzy C-means clustering algorithm, which combines the fuzzy membership and the geometric structure of the data set to achieve accurate and rapid classification of the dynamometer diagram, thereby obtaining accurate recognition results.
[0131] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0132] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0133] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.
[0134] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A dynamometer diagram diagnostic method, characterized in that: The specific steps include: S100, collecting measured dynamometer diagram data, obtaining polished rod displacement data as the abscissa, and obtaining suspension point load data at the polished rod suspension point as the ordinate, wherein the dynamometer diagram is a load-displacement dynamometer diagram; S200, normalizing the load-displacement dynamometer diagram, and converting the load-displacement dynamometer diagram into a 2-D dynamometer diagram; S300, extracting the time domain features of the 2-D dynamometer diagram as a first vector, and extracting the following time domain features: S311, obtain an average value, and use the average value as the first parameter, defined as ; S312. Obtain a quantile, and use the quantile as a second parameter, defined as ; S313. Obtain three means, and use the three means as the third parameter, defined as , where the three means are the weighted means of the upper quartile, median, and lower quartile; S314. Obtain a mean square value, and use the mean square value as a fourth parameter, defined as ; S315. Obtain the variance, and use the variance as the fifth parameter, defined as ; S316. Obtain the maximum amplitude, and use the maximum amplitude as the sixth parameter, defined as ; S317, obtaining a minimum amplitude, and using the minimum amplitude as the seventh parameter, defined as ; S318, obtaining a peak value difference, and using the peak value difference as an eighth parameter, defined as ; The time domain feature vector of the 2-D dynamometer diagram obtained after time domain feature extraction according to the first to eighth parameters is the first vector S: ； S400, obtaining the time-frequency domain features of the input signal as a second vector, S400 specifically includes: S410, using adaptive noise complete set empirical mode decomposition to process the load signal to obtain the kurtosis characteristics of the intrinsic mode function ; In the formula, ,in, is the kurtosis coefficient of the component signal of the first three intrinsic mode functions; S420, performing Hilbert transform on the intrinsic mode function to extract marginal spectrum frequency characteristics ; In the formula, is the centroid frequency in the marginal spectrum, is the mean square frequency in the marginal spectrum, is the frequency variance in the marginal spectrum, and the time-frequency domain features are obtained as the second vector X: ； S500, combining the first vector obtained in S300 and the second vector obtained in S400 to obtain a working condition vector, and inputting the working condition vector into an improved clustering algorithm to obtain a working condition classification result; The first vector S obtained in S300 and the second vector X obtained in S400 are combined to obtain a corresponding operating condition vector Z=[S, X], and the operating condition vector Z=[S, X] is input into the improved clustering algorithm to obtain an indicator diagram operating condition recognition model based on the improved clustering algorithm, and then the operating condition classification result is output, including: Normal working conditions, gas-affected working conditions, insufficient fluid supply conditions, reduction gearbox failure conditions, and sand production conditions; The improved clustering algorithm is an improved fuzzy C-means clustering algorithm, and the improved fuzzy C-means clustering algorithm includes: S1a, initialization, setting the initial parameters of the fuzzy C-means clustering algorithm; S2a, obtaining cluster centers, then obtaining data points corresponding to each cluster center, obtaining the membership degree corresponding to each data point according to the data points, and updating the cluster centers according to the membership degrees; S3a, obtaining evaluation parameters according to the cluster centers, data points, membership degrees, and the maximum eigenvalue and minimum eigenvalue of the cluster covariance matrix corresponding to each of the data points obtained after the update; establishing a cluster validity function according to the evaluation parameters, optimizing the cluster validity function to obtain a first optimal result, and then obtaining an optimal number of clusters corresponding to the first optimal result; Step S3a specifically includes the following steps: S3a1, according to the cluster center, data point, membership degree obtained after update, and the maximum eigenvalue and minimum eigenvalue of the cluster covariance matrix corresponding to each data point, respectively obtain: The compactness result, overlap result, cluster shape result, similarity result and separation result of the fuzzy C-means clustering algorithm are used to establish the evaluation parameters; S3a2, establishing a clustering validity function according to the evaluation parameters obtained in S3a1; The sum of the compactness result, the overlap result, and the cluster shape result is the first result M, and the sum of the similarity result and the separation result is the second result H; Get the clustering validity function F: F = M / H; The minimum value of the clustering validity function F min As the first optimal result, outputting the optimal number of clustering items corresponding to the first optimal result; S4a, updating the fuzzy C-means clustering algorithm with the optimal number of clusters to obtain an improved fuzzy C-means clustering algorithm.