Power diagram working condition analysis method based on dynamic clustering

Abstract

The invention provides an electric power diagram working condition analysis method based on dynamic clustering, and belongs to the technical field of oil pumping unit working condition diagnosis. The method comprises the following steps: acquiring actually measured electric power diagram data of the oil pumping unit, preprocessing the actually measured electric power diagram data, and drawing an electric power diagram according to the preprocessed electric power diagram data; respectively carrying out time domain analysis and frequency domain analysis on the electric power diagram, and extracting a time domain feature vector S and a frequency domain feature vector P from the electric power diagram; and combining the time domain feature vector S and the frequency domain feature vector P to obtain a working condition identification feature vector, carrying out dynamic clustering on the working condition identification feature vector by using a preset dynamic clustering method to obtain a diagnosis result of the working condition of the oil pumping unit, and constructing an electrical work diagram working condition identification model based on the preset dynamic clustering method. According to the method, the frequency domain analysis and the time domain analysis of the electrical work diagram are combined, the feature extraction method is enriched, the working condition recognition accuracy is improved, the technical bottleneck of electrical parameter working condition recognition is broken through, the engineering practicability is high, and the system research and development cost is low.

Description

Technical Field

[0001] The invention relates to the technical field of pumping unit operating condition diagnosis, and in particular to an electric power diagram operating condition analysis method based on dynamic clustering. Background Art

[0002] The widely used method for pumping unit working condition diagnosis is the diagnosis technology based on the dynamometer diagram, especially the analysis of the dynamometer diagram in combination with artificial intelligence methods to achieve working condition identification. Although this technology has shown relatively good results in oil well production, it still has significant limitations: 1. Insufficient real-time monitoring capabilities of data acquisition equipment: The collection of dynamometer data relies on high-cost equipment, and the installation of these devices often requires downtime operations, resulting in failures that cannot be discovered and handled in a timely manner, which may cause equipment damage or production losses, thereby interfering with daily production activities.

[0003] 2. Traditional diagnostic method manual analysis: Early dynamometer diagram analysis mainly relied on manual work, judging the downhole working conditions through experience and the characteristics of the dynamometer diagram. Although this method is effective to a certain extent, it is limited by the experience and skills of the analysts, and the accuracy and consistency of the diagnostic results are difficult to guarantee.

[0004] 3. Influence of damping coefficient: The damping coefficient is particularly critical to the analysis of the indicator diagram. Its inaccuracy will affect the accuracy of the characteristic parameters calculated from the pump indicator diagram, thereby reducing the effect of operating condition identification.

[0005] 4. The electrical parameters of the motor are easy to measure, with high data acquisition accuracy, good reliability and low cost. They are important parameters for long-term online monitoring of the pumping system and have obvious application advantages. Therefore, researchers have conducted a series of studies on the working condition diagnosis and identification technology based on the electrical power diagram. This working condition diagnosis technology has achieved certain results in oil well production, but it still has the following significant limitations: (1) Lack of intelligent and automated diagnostic models: Current electrical diagram fault diagnosis mostly uses traditional analysis methods and lacks intelligent diagnostic models based on machine learning and artificial intelligence, making it difficult to achieve automated fault identification and early warning.

[0006] (2) The current methods for extracting electrical parameter signal features are relatively simple, mainly based on mechanism analysis and expert experience: these methods identify the working conditions of the pumping unit by analyzing the electrical power diagram, which involves complex calculations of the characteristic values ​​of the electrical power diagram. In addition, the evaluation criteria for key characteristic values ​​are mainly based on human settings, which may cause the results to be affected by subjective factors.

[0007] (3) Different from general image recognition, oilfield electrical power diagrams are simple and have few feature points. Directly inputting traditional electrical power diagram images into the algorithm will, to a certain extent, lead to low feature utilization and large amount of data calculation.

[0008] (4) Adaptability and robustness need to be improved: The working environment of the oil pump is complex and changeable, and the characteristics of the electrical power diagram may change significantly under different working conditions. The existing diagnostic methods are insufficient in adaptability and robustness, which may lead to unstable diagnostic accuracy under different working conditions. Summary of the invention

[0009] In view of this, an embodiment of the present invention provides an electric power diagram working condition analysis method based on dynamic clustering, so as to solve the problems of low recognition accuracy and unstable diagnosis accuracy of existing electric power diagram working conditions.

[0010] The technical solution adopted by the present invention is: The present invention provides an electric power diagram working condition analysis method based on dynamic clustering, comprising: Collecting the measured electric power diagram data of the oil pumping unit, preprocessing the measured electric power diagram data, and drawing the electric power diagram based on the preprocessed electric power diagram data; Performing time domain analysis on the electric power diagram, and extracting a time domain feature vector S when the oil pump is working from the electric power diagram; Performing frequency domain analysis on the electric power diagram, and extracting a frequency domain feature vector P when the oil pump is working from the electric power diagram; The time domain feature vector S and the frequency domain feature vector P are combined to obtain the working condition identification feature vector, and the working condition identification feature vector is dynamically clustered using a preset dynamic clustering method to obtain the diagnosis result of the pumping unit working condition, and an electric power diagram working condition identification model is constructed based on the preset dynamic clustering method.

[0011] Further, the collecting of measured electric power diagram data of the oil pumping unit, preprocessing the measured electric power diagram data, and drawing an electric power diagram based on the preprocessed electric power diagram data includes: The electric power data and time data of the pumping well are obtained by using the SGT-2000 intelligent electric power meter installed at the suspension point of the pumping unit's polished rod; The electric power data and time data of the pumping well are cleaned, and the cleaned data are normalized. Based on the normalized electric power diagram data, a corresponding binary image is drawn to obtain the electric power diagram of the pumping unit.

[0012] Furthermore, the time domain analysis of the electric power diagram is performed to extract the time domain feature vector S when the oil pump is working from the electric power diagram, including: The electric power signal in the electric power diagram is analyzed in the time domain, and the average power value, root mean square value, variance, standard deviation, peak factor, mean waveform factor, kurtosis factor and skewness factor of the electric power signal are calculated respectively. The average power value, root mean square value, variance, standard deviation, peak factor, mean waveform factor, kurtosis factor and skewness factor are combined to obtain the time domain feature vector ; in, , Represents the average power value of the electric power signal in the entire sampling period, Indicates the total number of sampling points of the electric power signal; represents the electric power signal value collected at the nth time point; , Indicates the RMS value of the electric power signal; , represents the variance of the electric power signal; , Represents the standard deviation of the electric power signal; , Represents the crest factor of the electric power signal, Indicates the maximum value of the electric power signal in the entire sampling period; , Represents the mean form factor of the electric power signal; , Represents the kurtosis factor of the electric power signal; , Represents the skewness factor of the electric power signal.

[0013] Furthermore, the frequency domain analysis of the electric power diagram is performed to extract the frequency domain feature vector P when the oil pump is working from the electric power diagram, including: The electric power signal in the electric power diagram is analyzed in the frequency domain, and the vibration energy, spectrum variance, spectrum skewness, spectrum kurtosis, centroid frequency, spectrum standard deviation and band center of the electric power signal are calculated respectively. The vibration energy, spectrum variance, spectrum skewness, spectrum kurtosis, centroid frequency, spectrum standard deviation and band center are combined to obtain the frequency domain feature vector , in, , Represents the total energy of the electrical power signal; is the number of spectrum lines, which indicates the number of data points in the spectrum obtained by Fourier transform; Represents the electric power signal in the frequency domain A spectrum value, indicating the amplitude at that frequency; , represents the spectrum variance of the electric power signal; , Indicates the spectral skewness of the electric power signal; , Represents the spectral kurtosis of the electric power signal; , represents the centroid frequency of the electric power signal, is the frequency value of the spectrum line, indicating the frequency corresponding to the spectrum line; , Represents the standard deviation of the spectrum of the electric power signal; , Indicates the center of the frequency band of the electric power signal.

[0014] Furthermore, the time domain feature vector S and the frequency domain feature vector P are combined to obtain a working condition identification feature vector, and the working condition identification feature vector is dynamically clustered using a preset dynamic clustering method to obtain a diagnosis result of the pumping unit working condition, including: The time domain feature vector S and the frequency domain feature vector P are combined to obtain the working condition identification feature vector , the preset dynamic clustering method is used to dynamically cluster the working condition identification feature vectors, find the optimal clustering center, and interpret the working condition of the pumping unit based on the final dynamic clustering result.

[0015] Furthermore, the preset dynamic clustering method includes: Randomly generate an initial solution , as the starting point of the current local minimum, and set the continuous rejection counter , maximum number of consecutive rejections , used to determine whether to continue the local optimal search; allowable error , used to judge convergence; For the current local minimum , generates a random perturbation vector , so that the new starting point is: ; Random perturbation vector ,in, is a step size coefficient, is a random vector that follows a normal distribution; Use the sequential quadratic programming algorithm to find new candidate points Start a local optimal search to solve the local minimization problem and get a new local minimum ; Determining the local minimum Is it better than the current local minimum? , if the local minimum Better than the current local minimum , then the local minimum as a new local optimal solution; if the current local minimum Better than the local minimum Y, then keep the current local minimum is the current local optimal solution; The local minimum As a candidate solution for global optimization, it outputs the final local minimum when the global optimum is reached. as the global optimal solution.

[0016] Furthermore, the sequential quadratic programming algorithm is used to Start a local optimal search to solve the local minimization problem and get a new local minimum ,include: Use Taylor expansion to solve the objective function Make a quadratic approximation: in, The objective function is at point The gradient of is the Hessian matrix; The local minimum at the current iteration point Nearby, the iteration point is updated by solving the following quadratic programming problem: Iterate local optimization with constraints to find the local optimum. The constraints are as follows: in, As constraints, For x k The increment of Update the point of the iterative local minimum by solving the quadratic programming subproblem: , in, is the point of the updated local minimum.

[0017] Furthermore, the determination of the local minimum Is it better than the current local minimum? ,include: Comparing local minima and the current local minimum The objective function value of is: If the objective function value , then the local minimum As a new local optimal solution, , the consecutive rejection counter Reset to 0 and regenerate a new random perturbation vector for local optimization; If the objective function value , then reject the local minimum , and keep the current local optimal solution constant, , the consecutive rejection counter Increase by 1; Determine the consecutive rejection counter Whether the maximum allowed value has been reached ,if , then iteratively perform new random perturbations and local searches; if , then the random perturbation and local search are terminated.

[0018] In summary, the beneficial effects of the present invention are as follows: The present invention provides a method for analyzing the working condition of an electric power diagram based on dynamic clustering. The method extracts a time domain feature vector S and a frequency domain feature vector P from the electric power diagram by performing time domain analysis and frequency domain analysis on the electric power diagram of the oil pump, respectively. By combining the frequency domain and time domain analysis of the electric power diagram, the feature extraction method is enriched and the accuracy of working condition identification is improved. Finally, the time domain feature vector S and the frequency domain feature vector P are combined to obtain a working condition identification feature vector. The working condition identification feature vector is dynamically clustered using a preset dynamic clustering method to obtain a diagnosis result of the working condition of the oil pump. Based on the preset dynamic clustering method, an electric power diagram working condition identification model is constructed, which breaks through the technical bottleneck of electrical parameter working condition identification, has strong engineering practicality, and has low system development cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following is a brief introduction to the drawings required for use in the embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work, and these are all within the protection scope of the present invention.

[0020] Figure 1 It is a flow chart of a method for analyzing working conditions of an electric power diagram based on dynamic clustering of the present invention; Figure 2 This is a flow chart of the dynamic clustering analysis of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. If there is no conflict, the various features of the present invention and the embodiments can be combined with each other, all within the scope of protection of the present invention.

[0022] The working condition diagnosis method of the pumping unit began to use the electric power diagram for working condition diagnosis, and a series of studies were carried out on the working condition diagnosis and identification technology based on the electric power diagram. This working condition diagnosis technology has achieved certain results in oil well production, but it still has the following shortcomings: (1) Traditional electric power diagrams have the characteristics of simple images and few feature points. Directly inputting traditional electric power diagram images into the algorithm will, to a certain extent, lead to low feature utilization, large amount of data calculation, and low working condition recognition accuracy.

[0023] (2) The feature vectors extracted by the existing technology are usually relatively simple when performing feature extraction. However, the fault types of pumping wells are diverse, which are reflected not only in the time domain of the electric power diagram curve, but also in the frequency domain of the electric power diagram. Therefore, how to fuse and normalize multiple feature information and judge the working condition type more accurately and concisely is the key point of working condition diagnosis.

[0024] (3) Traditional sequential quadratic programming (SQP) clustering algorithms require a good initial trajectory to achieve relatively good optimization results. The algorithm has weak global search capabilities but strong local search capabilities, and can converge to local extreme points in a relatively short time. However, for objective functions with multiple local extreme points, SQP is very sensitive to the given initial points and is prone to falling into local extreme points and failing to obtain the global optimal value. The monotone basin hopping (MBH) algorithm can find a global optimal solution or a solution close to the optimal solution. However, when applied to clustering problems, there is still a risk of falling into a local optimal solution due to the existence of multiple local optimal solutions.

[0025] Therefore, considering that the traditional electric power diagram has the characteristics of simple image and few feature points, directly inputting the traditional electric power diagram image into the algorithm will, to a certain extent, lead to the problems of low feature utilization, large amount of data calculation, and low working condition recognition accuracy. The present invention combines the frequency domain and time domain analysis of the electric power diagram to extract the time domain feature vector and the frequency domain feature vector to form the working condition recognition vector. The feature extraction task is performed while compressing the data, breaking through the technical bottleneck of electrical parameter working condition recognition, with strong engineering practicality and low system development cost.

[0026] At the same time, existing supervised learning-based methods rely heavily on training samples. In order to achieve unsupervised learning of downhole fault diagnosis of pumping units, the present invention proposes a dynamic clustering technology. The determination of the cluster center is a very important parameter that affects the clustering accuracy. It is difficult to accurately set it in advance. It is regarded as a variable parameter and dynamically adjusted during the clustering process. The present invention combines the monotone basin jumping algorithm with the nonlinear programming algorithm to give full play to the advantages of both and make up for each other's shortcomings. The monotone basin jumping algorithm provides diversity and global search capabilities, while the nonlinear programming algorithm provides local optimization capabilities and faster convergence speed. This combined method can improve the performance of the optimization algorithm in non-convex optimization problems and provide a more reliable and efficient solution. The detailed implementation process of the present invention is shown in the following embodiments.

[0027] Example 1: Reference Figure 1 As shown, Figure 1 This is a flow chart of a method for analyzing the working condition of an electric power diagram based on dynamic clustering of the present invention. Figure 1 As shown, the method for analyzing the working condition of the electric power diagram based on dynamic clustering of the present invention comprises: Collecting the measured electric power diagram data of the oil pumping unit, preprocessing the measured electric power diagram data, and drawing the electric power diagram based on the preprocessed electric power diagram data; Performing time domain analysis on the electric power diagram, and extracting a time domain feature vector S when the oil pump is working from the electric power diagram; Performing frequency domain analysis on the electric power diagram, and extracting a frequency domain feature vector P when the oil pump is working from the electric power diagram; The time domain feature vector S and the frequency domain feature vector P are combined to obtain the working condition identification feature vector, and the working condition identification feature vector is dynamically clustered using a preset dynamic clustering method to obtain the diagnosis result of the pumping unit working condition, and an electric power diagram working condition identification model is constructed based on the preset dynamic clustering method.

[0028] Specifically, in the embodiment of the present invention, the measured electric power diagram data of the oil pumping unit is collected, the measured electric power diagram data is preprocessed, and the electric power diagram is drawn according to the preprocessed electric power diagram data, which specifically includes: The electric power data and time data of the pumping well are obtained using the SGT-2000 intelligent electric power meter installed at the suspension point of the pumping unit's polished rod.

[0029] The electric power data and time data of the pumping well are cleaned, and the cleaned data are normalized. Based on the normalized electric power diagram data, a corresponding binary image is drawn to obtain the electric power diagram of the pumping unit.

[0030] The present invention removes abnormal values, wrong values, and null values ​​through data cleaning to provide accurate and complete data for subsequent research, and stores the cleaned data for subsequent query. The electric power diagram is a binary image drawn with time as the horizontal axis and power as the vertical axis.

[0031] Furthermore, in an embodiment of the present invention, a time domain analysis is performed on the electric power diagram, and a time domain feature vector S when the oil pump is working is extracted from the electric power diagram, including: The electric power signal in the electric power diagram is analyzed in the time domain, and the average power value, root mean square value, variance, standard deviation, peak factor, mean waveform factor, kurtosis factor and skewness factor of the electric power signal are calculated respectively. The average power value, root mean square value, variance, standard deviation, peak factor, mean waveform factor, kurtosis factor and skewness factor are combined to obtain the time domain feature vector: , in, , Represents the average power value of the electric power signal in the entire sampling period, Indicates the total number of sampling points of the electric power signal; Represents the electric power signal value collected at the nth time point.

[0032] , It represents the RMS value of the electric power signal. The RMS value is a measure of the energy of the electric power signal. This parameter reflects the energy level or amplitude of the signal. The higher the value, the greater the vibration amplitude of the signal.

[0033] , Indicates the variance of the electric power signal. The variance indicates the degree of dispersion of the electric power signal relative to the mean, and is obtained by calculating the square difference between the electric power signal and the mean. The variance reflects the degree of fluctuation or dispersion of the electric power signal. The larger the value, the more drastic the fluctuation of the electric power signal.

[0034] , Indicates the standard deviation of the electric power signal. The standard deviation is the square root of the variance and indicates the degree of dispersion of the electric power signal. This parameter is used to quantify the fluctuation range of the signal value and is a measure of signal stability.

[0035] , Represents the crest factor of the electric power signal, It indicates the maximum value of the electric power signal in the entire sampling period. This parameter is used to measure the sharpness or impact of the electric power signal. The larger the peak factor, the stronger the instantaneous impact in the electric power signal.

[0036] , Indicates the mean waveform factor of the electric power signal. This parameter is used to quantify the shape of the electric power signal and reflects the waveform characteristics of the electric power signal.

[0037] , Indicates the kurtosis factor of the electric power signal. This factor is used to describe the peak characteristics of the electric power signal. The higher the value, the more obvious the peak.

[0038] , Represents the skewness factor of the electric power signal. The skewness factor is used to analyze the symmetry of the electric power signal. A large value indicates significant signal asymmetry.

[0039] Furthermore, in an embodiment of the present invention, a frequency domain analysis is performed on the electric power diagram, and a frequency domain feature vector P when the oil pump is working is extracted from the electric power diagram, including: The electric power signal in the electric power diagram is analyzed in the frequency domain, and the vibration energy, spectrum variance, spectrum skewness, spectrum kurtosis, centroid frequency, spectrum standard deviation and band center of the electric power signal are calculated respectively. The vibration energy, spectrum variance, spectrum skewness, spectrum kurtosis, centroid frequency, spectrum standard deviation and band center are combined to obtain the frequency domain feature vector , Where, represents the total energy of the electric power signal; , Represents the total energy of the electric power signal and is used to evaluate the vibration intensity of the electric power signal; is the number of spectrum lines, which indicates the number of data points in the spectrum obtained by Fourier transform; Represents the electric power signal in the frequency domain A spectrum value, indicating the amplitude at that frequency.

[0040] , Indicates the spectrum variance of the electric power signal. The spectrum variance describes the degree of diffusion of the spectrum distribution. The higher the value, the wider the spectrum distribution.

[0041] , Indicates the spectral skewness of the electric power signal. The spectral skewness describes the symmetry of the spectrum. The larger the value, the more significant the asymmetry of the spectrum.

[0042] , Indicates the spectral kurtosis of the electric power signal. The spectral kurtosis characterizes the degree of "peak" of the spectrum. The higher the value, the more peaks there are in the spectrum.

[0043] , Indicates the centroid frequency of the electric power signal. The centroid frequency indicates the center frequency of the spectrum and is used to characterize the center position of the spectrum distribution. is the frequency value of the spectrum line, indicating the frequency corresponding to the spectrum line.

[0044] , It represents the standard deviation of the spectrum of the electric power signal, describes the degree of dispersion of the spectrum, and is used to measure the width of the spectrum distribution.

[0045] , Indicates the frequency band center of the electric power signal. Calculates the center frequency position of the spectrum to locate the main frequency band.

[0046] Further, in an embodiment of the present invention, the time domain feature vector S and the frequency domain feature vector P are combined to obtain a working condition identification feature vector, and the working condition identification feature vector is dynamically clustered using a preset dynamic clustering method to obtain a diagnosis result of the pumping unit working condition, including: The time domain feature vector S and the frequency domain feature vector P are combined to obtain the working condition identification feature vector , the preset dynamic clustering method is used to dynamically cluster the working condition identification feature vectors, find the optimal clustering center, and interpret the working condition of the pumping unit based on the final dynamic clustering result.

[0047] Furthermore, in the embodiment of the present invention, the monotone basin hopping (MBH) algorithm has good global search capability and good robustness, and the sequential quadratic programming (SQP) algorithm is currently recognized as the most effective method for solving constrained nonlinear optimization problems. The present invention proposes a novel clustering algorithm based on monotone basin hopping (MBH) + sequential quadratic programming (SQP). Figure 2 As shown, the present invention provides a comprehensive optimization method of global and local search, that is, a preset dynamic clustering method, and its detailed steps are as follows: Initialization, randomly generate an initial solution , as the starting point of the current local minimum, this point is evenly distributed in the entire search space to ensure the diversity of the initial points. At the same time, set the continuous rejection counter , maximum number of consecutive rejections , used to determine whether to continue the local optimal search; allowable error , which is used to judge convergence.

[0048] For the current local minimum , generates a random perturbation vector , R is a set of real numbers; let the new starting point be: , the perturbation vector is used to jump out of the area where the local minimum value is located to increase the ability of global search.

[0049] Random perturbation vector ,in, is a step size coefficient, is a random vector that follows a normal distribution; Use the sequential quadratic programming algorithm (SQP algorithm) to find new candidate points Start a local optimal search to solve the local minimization problem and get a new local minimum SQP first uses Taylor expansion to perform a quadratic approximation on the objective function and optimizes it in combination with constraints to find the local optimum. In each iteration, the objective function and constraints to solve the local optimization problem.

[0050] Determining the local minimum Is it better than the current local minimum? , if the local minimum Better than the current local minimum , then the local minimum as a new local optimal solution; if the current local minimum Better than the local minimum Y, then keep the current local minimum is the current local optimal solution; The local minimum As a candidate solution for global optimization, the "bottom of the funnel" solution is considered to have reached the vicinity of the global optimum, the algorithm is terminated, and the final local minimum is output when the global optimum is reached. as the global optimal solution.

[0051] Furthermore, the sequential quadratic programming algorithm is used to Start a local optimal search to solve the local minimization problem and get a new local minimum ,include: Use Taylor expansion to solve the objective function Make a quadratic approximation: in, The objective function is at point The gradient of is the Hessian matrix; The local minimum at the current iteration point Nearby, the iteration point is updated by solving the following quadratic programming problem: Iterate local optimization with constraints to find the local optimum. The constraints are as follows: in, As constraints, For x k The increment of Update the point of the iterative local minimum by solving the quadratic programming subproblem: , is the point of the updated local minimum.

[0052] Furthermore, the determination of the local minimum Is it better than the current local minimum? ,include: Comparing local minima and the current local minimum The objective function value of is: If the objective function value , then the local minimum As a new local optimal solution, , the consecutive rejection counter Reset to 0 and regenerate a new random perturbation vector for local optimization; If the objective function value , then reject the local minimum , and keep the current local optimal solution constant, , the consecutive rejection counter Increase by 1; Determine the consecutive rejection counter Whether the maximum allowed value has been reached ,if , then iteratively perform new random perturbations and local searches; if , it is considered that the current solution cannot be further improved, and the loop is jumped out, the random perturbation and local search are ended, and the termination step is entered.

[0053] The present invention has the following technical effects: (1) Completely detaching from the dynamometer diagram for operating condition diagnosis, since the oil pump works thousands of meters underground, the operating conditions are very complex, the working environment is extremely harsh, and the failure rate is very high. Therefore, the dynamometer diagram displayed on site cannot accurately reflect the working condition of the oil pump. The electrical parameters of the motor are easy to measure, and the data acquisition is accurate, reliable, and low-cost. They are important parameters for long-term online monitoring of the oil pump system and have obvious application advantages. For a long time, due to the lack of an applicable overall model and an effective feature extraction method, the evaluation method of the oil pump based on the dynamometer diagram has not been studied in depth. This paper proposes a dynamometer diagram operating condition recognition model using dynamic clustering technology.

[0054] (2) By combining the frequency domain and time domain analysis of the electric power diagram, the feature extraction method is enriched and the accuracy of working condition identification is improved. The evaluation criteria of key characteristic values ​​are obtained by computer calculation, which eliminates the influence of subjective factors on the diagnosis results and ensures the identification accuracy. It breaks through the technical bottleneck of electrical parameter working condition identification, has strong engineering practicality, and low system R&D cost.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing working conditions of electric power diagram based on dynamic clustering, characterized in that: include: Collecting the measured electric power diagram data of the oil pumping unit, preprocessing the measured electric power diagram data, and drawing the electric power diagram based on the preprocessed electric power diagram data; Performing time domain analysis on the electric power diagram, and extracting a time domain feature vector S when the oil pump is working from the electric power diagram; Performing frequency domain analysis on the electric power diagram, and extracting a frequency domain feature vector P when the oil pump is working from the electric power diagram; The time domain feature vector S and the frequency domain feature vector P are combined to obtain the working condition identification feature vector, and the working condition identification feature vector is dynamically clustered using a preset dynamic clustering method to obtain the diagnosis result of the pumping unit working condition, and an electric power diagram working condition identification model is constructed based on the preset dynamic clustering method.

2. The method for analyzing the working condition of an electric power diagram based on dynamic clustering according to claim 1, characterized in that: The collecting of measured electric power diagram data of the oil pumping unit, preprocessing the measured electric power diagram data, and drawing an electric power diagram based on the preprocessed electric power diagram data, includes: The electric power data and time data of the pumping well are obtained by using the SGT-2000 intelligent electric power meter installed at the suspension point of the pumping unit's polished rod; The electric power data and time data of the pumping well are cleaned, and the cleaned data are normalized. Based on the normalized electric power diagram data, a corresponding binary image is drawn to obtain the electric power diagram of the pumping unit.

3. The method for analyzing the working condition of an electric power diagram based on dynamic clustering according to claim 1, characterized in that: The time domain analysis of the electric power diagram is performed to extract the time domain feature vector S when the oil pump is working from the electric power diagram, including: The electric power signal in the electric power diagram is analyzed in the time domain, and the average power value, root mean square value, variance, standard deviation, peak factor, mean waveform factor, kurtosis factor and skewness factor of the electric power signal are calculated respectively. The average power value, root mean square value, variance, standard deviation, peak factor, mean waveform factor, kurtosis factor and skewness factor are combined to obtain the time domain feature vector: , in, , Represents the average power value of the electric power signal in the entire sampling period, Indicates the total number of sampling points of the electric power signal; represents the electric power signal value collected at the nth time point; , Indicates the RMS value of the electric power signal; , represents the variance of the electric power signal; , Represents the standard deviation of the electric power signal; , Represents the crest factor of the electric power signal, Indicates the maximum value of the electric power signal in the entire sampling period; , Represents the mean form factor of the electric power signal; , Represents the kurtosis factor of the electric power signal; , Represents the skewness factor of the electric power signal.

4. The method for analyzing the working condition of an electric power diagram based on dynamic clustering according to claim 1, characterized in that: The frequency domain analysis of the electric power diagram is performed to extract the frequency domain feature vector P when the oil pump is working from the electric power diagram, including: The electric power signal in the electric power diagram is analyzed in the frequency domain, and the vibration energy, spectrum variance, spectrum skewness, spectrum kurtosis, centroid frequency, spectrum standard deviation and band center of the electric power signal are calculated respectively. The vibration energy, spectrum variance, spectrum skewness, spectrum kurtosis, centroid frequency, spectrum standard deviation and band center are combined to obtain the frequency domain feature vector ,in, , Represents the total energy of the electrical power signal; is the number of spectrum lines, which indicates the number of data points in the spectrum obtained by Fourier transform; Represents the electric power signal in the frequency domain A spectrum value, indicating the amplitude at that frequency; , represents the spectrum variance of the electric power signal; , Indicates the spectral skewness of the electric power signal; , Represents the spectral kurtosis of the electric power signal; , represents the centroid frequency of the electric power signal, is the frequency value of the spectrum line, indicating the frequency corresponding to the spectrum line; , Represents the standard deviation of the spectrum of the electric power signal; , Indicates the center of the frequency band of the electric power signal.

5. The method for analyzing the working condition of an electric power diagram based on dynamic clustering according to claim 1, characterized in that: The time domain feature vector S and the frequency domain feature vector P are combined to obtain a working condition identification feature vector, and a preset dynamic clustering method is used to dynamically cluster the working condition identification feature vector to obtain a diagnosis result of the pumping unit working condition, including: The time domain feature vector S and the frequency domain feature vector P are combined to obtain the working condition identification feature vector , the preset dynamic clustering method is used to dynamically cluster the working condition identification feature vectors, find the optimal clustering center, and interpret the working condition of the pumping unit based on the final dynamic clustering result.

6. The method for analyzing the working condition of an electric power diagram based on dynamic clustering according to claim 1, characterized in that: The preset dynamic clustering method includes: Randomly generate an initial solution , as the starting point of the current local minimum, and set the continuous rejection counter , maximum number of consecutive rejections , used to determine whether to continue the local optimal search; allowable error , used to judge convergence; For the current local minimum , generates a random perturbation vector , so that the new starting point is: ; Random perturbation vector ,in, is a step size coefficient, is a random vector that follows a normal distribution; Use the sequential quadratic programming algorithm to find new candidate points Start a local optimal search to solve the local minimization problem and get a new local minimum ; Determining the local minimum Is it better than the current local minimum? , if the local minimum Better than the current local minimum , then the local minimum as a new local optimal solution; if the current local minimum Better than the local minimum Y, then keep the current local minimum is the current local optimal solution; The local minimum As a candidate solution for global optimization, it outputs the final local minimum when the global optimum is reached. as the global optimal solution.

7. The method for analyzing the working condition of an electric power diagram based on dynamic clustering according to claim 6, characterized in that: The new candidate points are generated by using a sequential quadratic programming algorithm. Start a local optimal search to solve the local minimization problem and get a new local minimum ,include: Use Taylor expansion to solve the objective function Make a quadratic approximation: in, The objective function is at point The gradient of is the Hessian matrix; The local minimum at the current iteration point Nearby, the iteration point is updated by solving the following quadratic programming problem: Iterate local optimization with constraints to find the local optimum. The constraints are as follows: in, As constraints, For x k The increment of Update the point of the iterative local minimum by solving the quadratic programming subproblem: in, is the point of the updated local minimum.

8. The method for analyzing the working condition of an electric power diagram based on dynamic clustering according to claim 6, characterized in that: Determining the local minimum Is it better than the current local minimum? ,include: Comparing local minima and the current local minimum The objective function value of is: If the objective function value , then the local minimum As a new local optimal solution, , the consecutive rejection counter Reset to 0 and regenerate a new random perturbation vector for local optimization; If the objective function value , then reject the local minimum , and keep the current local optimal solution constant, , the consecutive rejection counter Increase by 1; Determine the consecutive rejection counter Whether the maximum allowed value has been reached ,if , then iteratively perform new random perturbations and local searches; if , then the random perturbation and local search are terminated.

Images