Polish rod rotator operation state monitoring method based on multi-source sensor data

Abstract

The invention provides a polish rod rotator operation state monitoring method and system based on multi-source sensor data, and relates to the technical field of data processing, and the method comprises the steps: obtaining reed switch pull-in data; judging whether the polish rod rotator is abnormal or not according to the pull-in data of the reed switch; if yes, entering the next step, and activating operation state monitoring of the oil pumping unit; acquiring acceleration sensor data; judging whether the oil pumping unit operates normally or not according to acceleration change characteristics of data of the acceleration sensor; if yes, entering the next step, and activating polish rod rotator operation state monitoring; according to data of the acceleration sensor, stroke frequency data of the oil pumping unit are calculated; waiting for the next time of pull-in data of the reed switch; calculating the rotation angle of the polish rod rotator according to the stroke frequency data of the oil pumping unit and the interval time between the two times of pull-in of the reed switch; judging whether the polish rod rotator is abnormal or not according to the rotation angle of the polish rod rotator; and if yes, outputting a fault alarm of the polish rod rotator.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a method and system for monitoring the running state of a polished rod rotary based on multi-source sensor data. BACKGROUND

[0002] With the increasing demand for energy, the production efficiency and reliability of equipment in the oil industry have become increasingly important. As a key facility in the oil production process, the polished rod rotary, its running state monitoring is of great significance to ensure production safety, improve equipment utilization, reduce downtime and maintenance costs.

[0003] With the rapid development of science and technology, and the urgent need for monitoring the running state of the polished rod rotary, more and more modern technologies have been applied to the monitoring of the running state of the polished rod rotary. There are many modern monitoring technologies based on sensors applied to the monitoring of the polished rod rotary.

[0004] However, the current state monitoring method mainly relies on sensors, which need to work continuously and collect data all day long. The continuously running sensors will generate a large amount of data flow, and processing these data will not only increase the burden of the system, but also may lead to waste of resources and decrease of response speed. SUMMARY

[0005] In view of the above deficiencies of the prior art, the purpose of the embodiments of the present application is to provide a method for monitoring the running state of a polished rod rotary based on multi-source sensor data, which can solve the technical problems that the continuously running sensors will generate a large amount of data flow, and processing these data will not only increase the burden of the system, but also may lead to waste of resources and decrease of response speed.

[0006] The first aspect of the embodiments of the present application proposes a method for monitoring the running state of a polished rod rotary based on multi-source sensor data, comprising: S1: obtaining dry reed suction data; S2: determining whether the polished rod rotary is abnormal according to the dry reed suction data; if yes, activating the running state monitoring of the pumping unit, otherwise, maintaining dormant and returning to S1 for continuous monitoring; S3: obtaining acceleration sensor data; S4: determining whether the pumping unit is running normally according to the acceleration change characteristics of the acceleration sensor data; if yes, activating the running state monitoring of the polished rod rotary, otherwise, returning to S3; S5: calculating pumping unit stroke data according to the acceleration sensor data; S6: waiting for the next dry reed suction data; S7: calculating a rotating angle of the polished rod rotary according to the pumping unit stroke data and interval time between two times of dry reed attraction; S8: judging whether the polished rod rotary is abnormal according to the rotating angle of the polished rod rotary; if yes, outputting a polished rod rotary fault alarm.

[0007] The second aspect of the embodiment of the present application provides a polished rod rotary running state monitoring system based on multi-source sensor data, comprising a processor and a memory. The memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the steps of the polished rod rotary running state monitoring method based on multi-source sensor data as described in the first aspect.

[0008] The third aspect of the embodiment of the present application provides a readable storage medium, and the readable storage medium stores programs or instructions, and the programs or instructions are executed by the processor to implement the steps of the polished rod rotary running state monitoring method based on multi-source sensor data as described in the first aspect.

[0009] The technical scheme provided by the embodiment of the present application has at least the following beneficial effects: In the embodiment of the present application, whether to activate subsequent sensor monitoring is determined through preliminary abnormality detection of dry reed attraction data, and through intelligent hibernation and activation mechanism, the system will further activate the monitoring of the acceleration sensor and the pumping unit stroke data only when the dry reed data is abnormal, thereby effectively reducing unnecessary calculation burden and resource consumption. BRIEF DESCRIPTION OF DRAWINGS

[0010] The accompanying drawings are only for the purpose of illustrating specific embodiments and are not considered as limiting the present application. In the entire drawings, the same reference signs represent the same components. Obviously, the accompanying drawings in the following description are only some embodiments described in the embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0011] Figure 1 is a flowchart of a polished rod rotary running state monitoring method based on multi-source sensor data provided by the embodiment of the present application.

[0012] Figure 2 is a structural diagram of a polished rod rotary running state monitoring system based on multi-source sensor data provided by the embodiment of the present application. DETAILED DESCRIPTION

[0013] In order for those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions of the present application will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. It should be understood that these descriptions are only exemplary, but not used to limit the scope of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0014] The method for monitoring the running state of the polished rod rotary based on multi-source sensor data provided by the embodiments of the present application will be described in detail below with reference to the drawings, specific embodiments and application scenarios.

[0015] Referring to the drawings attached Figure 1 , a flowchart of a method for monitoring the running state of the polished rod rotary based on multi-source sensor data provided by the embodiments of the present application is shown.

[0016] The embodiments of the present application provide a method for monitoring the running state of the polished rod rotary based on multi-source sensor data, which can include the following steps: S1: Obtain dry reed pipe attraction data.

[0017] Among them, the dry reed pipe attraction data refers to the signal data obtained by the dry reed pipe sensor, which is a kind of switch sensor based on the principle of magnetic field change, usually used to monitor the start-stop state of the equipment. In the monitoring of the polished rod rotary, the dry reed pipe attraction data is usually used to detect the on-off state of the equipment, especially the working state of the pumping unit. When the dry reed pipe senses the change of the magnetic field, it will be attracted or disconnected, thereby outputting the corresponding binary signal (usually "0" represents not attracted, "1" represents attracted). This data has the characteristics of binary, which can reflect whether the equipment is in working state or non-working state. By analyzing the change of the dry reed pipe attraction data, early signals can be provided for the state monitoring and fault warning of the polished rod rotary equipment.

[0018] S2: According to the dry reed pipe attraction data, it is judged whether the polished rod rotary exists abnormity. If yes, the next step is entered, and the pumping unit running state monitoring is activated. Otherwise, it is maintained dormant, and returns to S1 to continue monitoring.

[0019] In one possible implementation, S2 is specifically: according to the comparison result of the dry reed pipe attraction time length in the dry reed pipe attraction data and the preset time length, it is judged whether the polished rod rotary exists abnormity. If the dry reed pipe attraction time length in the dry reed pipe attraction data is greater than the preset time length, the next step is entered, and the pumping unit running state monitoring is activated. Otherwise, it is maintained dormant, and returns to S1 to continue monitoring.

[0020] Wherein, the person skilled in the art can set the size of the preset time length according to the actual situation, and the application does not make any limitation.

[0021] In the embodiment of the application, the preliminary anomaly detection of the dry reed valve data is used to determine whether to activate the subsequent sensor monitoring. Through the intelligent sleep and activation mechanism, the system will further activate the monitoring of the acceleration sensor and the pumping unit stroke data only when the dry reed valve data is abnormal, thereby effectively reducing unnecessary calculation burden and resource consumption.

[0022] S3: Obtain acceleration sensor data.

[0023] S4: Determine whether the pumping unit is running normally according to the acceleration variation characteristics of the acceleration sensor data. If yes, proceed to the next step and activate the light rod rotator running state monitoring. Otherwise, return to S3.

[0024] In one possible implementation, S4 is specifically: determining whether the pumping unit is running normally according to a comparison result of a peak-to-peak value of the acceleration sensor data and a preset peak-to-peak value. If the peak-to-peak value of the acceleration sensor data is greater than the preset peak-to-peak value, proceed to the next step and activate the light rod rotator running state monitoring. Otherwise, return to S3.

[0025] Wherein, the peak-to-peak value (Peak-to-Peak Value, abbreviated as P-P value) refers to the difference between the maximum value and the minimum value in a signal, which is used to measure the amplitude range of the signal. The peak-to-peak value can reflect the amplitude variation of the object motion or vibration, and is of great significance for detecting the working state and fault mode of the equipment. When the equipment is abnormal, the vibration amplitude may increase, resulting in the peak-to-peak value exceeding the normal range, which becomes the basis for fault warning.

[0026] Wherein, the person skilled in the art can set the size of the preset peak-to-peak value according to the actual situation, and the application does not make any limitation.

[0027] In the embodiment of the application, the acceleration variation characteristics of the acceleration sensor data are used to determine whether the pumping unit is running normally. This method can monitor the working state of the pumping unit in real time and discover potential abnormalities or faults in a timely manner. For example, when the pumping unit fails, its vibration characteristics will change, showing changes in the fluctuation amplitude or frequency of the acceleration signal. In this way, the fault can be quickly identified and immediate measures can be taken to avoid the expansion of the fault. In addition, only when it is confirmed that the pumping unit is running normally, the light rod rotator running state monitoring is activated, realizing a hierarchical and step-by-step monitoring strategy, effectively saving calculation resources and energy consumption, and improving the accuracy and response speed of the monitoring.

[0028] S5: Calculate the pumping unit stroke data according to the acceleration sensor data.

[0029] Specifically, by identifying the wave crest and wave trough in the acceleration signal, the start and end positions of each cycle are determined, so as to calculate the vibration change corresponding to one cycle, and the cycle is one pumping stroke. The number of cycles occurring in a unit of time is counted to obtain the pumping stroke data of the pumping unit, reflecting the working frequency of the pumping unit per minute or per hour.

[0030] S6: waiting for the next dry reed attraction data.

[0031] S7: calculating the rotating angle of the polished rod rotator according to the pumping stroke data of the pumping unit and the interval time between two dry reed attractions.

[0032] In a possible implementation, S7 specifically comprises: calculating the rotating angle of the polished rod rotator according to the pumping stroke data of the pumping unit and the interval time between two dry reed attractions. ; Wherein, θ represents the rotating angle of the polished rod rotator, T represents the interval time between two dry reed attractions, and SPM represents the pumping stroke data of the pumping unit.

[0033] Specifically, first, the mechanical transmission ratio between the pumping unit and the polished rod rotator needs to be understood, that is, the rotating angle of the polished rod rotator corresponding to each pumping stroke of the pumping unit. According to the pumping stroke data of the pumping unit, the rotating angle of the polished rod rotator is calculated by multiplying the transmission ratio. The total rotating angle of the polished rod rotator at a certain moment is obtained by accumulating the calculated rotating angle.

[0034] S8: judging whether the polished rod rotator is abnormal according to the rotating angle of the polished rod rotator. If yes, outputting the polished rod rotator fault alarm.

[0035] In a possible implementation, S8 specifically comprises: judging whether the polished rod rotator is abnormal according to the comparison result of the rotating angle of the polished rod rotator and the preset rotating angle. If the rotating angle of the polished rod rotator is less than the preset rotating angle, outputting the polished rod rotator fault alarm.

[0036] Wherein, the person skilled in the art can set the size of the preset rotating angle according to the actual situation, which is not limited in the present application.

[0037] In the embodiment of the present application, the rotation angle of the polished rod rotary is used to determine whether the polished rod rotary is running normally. This method can accurately monitor the working state of the polished rod rotary. The rotation angle of the polished rod rotary reflects its regularity and stability. If the rotation angle of the polished rod rotary is abnormal or deviates from the normal range, it usually means that there is a mechanical failure or abnormality, such as jamming, excessive friction, etc. By monitoring the rotation angle of the polished rod rotary in real time, the device abnormalities can be found in time and the fault alarm can be triggered, avoiding the accumulation of faults leading to more serious equipment damage. Only when the polished rod rotary is running normally, the next step of deep state monitoring is entered, which further improves the monitoring efficiency, reduces unnecessary calculation burden, and ensures the accuracy and real-time response capability of the monitoring system.

[0038] In a possible implementation, the polished rod rotary running state monitoring method based on multi-source sensor data further comprises: S9: When the polished rod rotary does not have an abnormality, activate deep state monitoring, and determine whether the polished rod rotary is running normally according to the reed switch attraction data, acceleration sensor data and pumping unit stroke data by using a deep learning algorithm based on a progressive fusion mechanism. If yes, return to S1 for continuous monitoring. Otherwise, output a polished rod rotary fault alarm.

[0039] Optionally, in addition to the reed switch attraction data, acceleration sensor data and pumping unit stroke data, pressure sensor data, temperature sensor data, current sensor data and liquid level sensor data and other modal data can also be introduced to jointly realize the polished rod rotary running state monitoring.

[0040] In a possible implementation, the deep learning algorithm based on a progressive fusion mechanism is used to determine whether the polished rod rotary is running normally, which specifically includes sub-steps S901 to S904: S901: According to the data characteristics of the reed switch attraction data, acceleration sensor data and pumping unit stroke data, select the corresponding conversion mode to convert the reed switch attraction data, acceleration sensor data and pumping unit stroke data into two-dimensional image data.

[0041] It should be noted that the neural network has better feature extraction effect on two-dimensional image data than one-dimensional sensor data, because the structure such as convolutional neural network (CNN) is naturally suitable for capturing spatial structure features. The two-dimensional image contains spatial correlation information distributed along the horizontal and vertical directions at the same time, and the network can extract multi-level features such as edges, textures and shapes in the local area through two-dimensional convolution kernels, and combine them layer by layer to form high-level semantic representation. While one-dimensional sensor data is only continuous in the time dimension, it lacks spatial local correlation, and the network can extract limited types and levels of features. Therefore, converting sensor time series signals into two-dimensional image data can fully utilize the advantages of deep neural networks in spatial feature extraction and obtain more rich and discriminative feature representation. For this reason, the sensor time series signals are converted into two-dimensional image data for further feature extraction.

[0042] Optionally, S901 specifically includes S9011 to S9013: S9011: According to the data characteristics that the reed switch attraction data only has two states of attraction and non-attraction, a binary image conversion method is used to convert the reed switch attraction data into two-dimensional reed switch attraction image data.

[0043] Among them, the binary image conversion method is to map each element in the data to two discrete values (usually 0 and 1) according to a preset threshold, and convert it into image data. When processing binary state data (such as "attraction" and "non-attraction"), this method maps the state of each time point to a pixel value in the image (for example, 0 represents white and 1 represents black). This conversion method is simple and intuitive, which can effectively preserve the state change characteristics of the data and facilitate subsequent pattern recognition and feature extraction using image processing techniques (such as convolutional neural networks).

[0044] It should be noted that the reed switch attraction data itself is binary, only containing two states of attraction and non-attraction, and converting it into two-dimensional image data through the binary image conversion method can preserve its state switching characteristics with very low information loss, and clearly show the distribution of the attraction pattern in the time dimension, which facilitates the neural network to extract the regularity of the switch state and improve the sensitivity to abnormal start-stop state.

[0045] S9012: According to the data characteristics that the acceleration sensor data has time sequence dependence, a Markov transition field image conversion algorithm is used to convert the acceleration sensor data into two-dimensional acceleration sensor image data.

[0046] The Markov transition field image conversion algorithm is an image conversion method based on a Markov process, which converts time series data into a two-dimensional image using a state transition matrix. In this method, continuous data is first discretized into multiple states, and then the transition probabilities between states are calculated to form a transition matrix. Finally, these probabilities are mapped into a two-dimensional image to capture the temporal dependence and state transition rules in the time series data. This algorithm can effectively reflect the evolution of signals over time and is suitable for processing sensor data with temporal correlation.

[0047] It should be noted that the acceleration sensor data has significant temporal dependence. By converting it into a two-dimensional image using the Markov transition field image conversion algorithm, the transition probability relationship between different states in the signal can be preserved, and the dynamic patterns in the time series can be expressed in a spatial form, allowing the neural network to better capture the evolution of the vibration pattern and improve the recognition accuracy of abnormal mechanical vibrations.

[0048] Optionally, S9012 specifically includes S90121 to S90125: S90121: Box processing the one-dimensional acceleration sensor data according to quantiles to obtain multiple states.

[0049] Specifically, the step of box processing the one-dimensional acceleration sensor data according to quantiles first sorts the acceleration data, and then divides the data into several intervals according to the pre-set quantiles. For example, if the data is divided into Q states, the data can be divided into Q equal frequency intervals, each containing an equal number of data points. The specific operation is: first calculate the quantile points of the data set (for example, the value corresponding to the Q quantile), then divide the data points according to these quantiles so that the value range in each interval is equal or the number is equal. Finally, the data points are mapped to the corresponding states according to the interval they belong to, and each state represents a certain specific range of acceleration data. Such box processing not only converts continuous acceleration signals into a limited number of discrete states, but also reduces the influence of noise to some extent and helps capture important change features in the data.

[0050] It should be noted that by box processing and calculating transition probabilities, the temporal dependence in the acceleration sensor data, i.e. the influence of the current state on the future state, can be effectively modeled. This provides more comprehensive information for analyzing the trend, pattern and potential faults of the vibration signal.

[0051] S90122: Calculate the transition probabilities between all states by maximum likelihood estimation.

[0052] Among them, Maximum Likelihood Estimation (MLE) is a method for estimating the parameters of a statistical model, the core idea of which is: given the observed data, select a set of parameters that maximize the probability (likelihood) of generating these observations under that parameter.

[0053] S90123: Normalize the transition probability to obtain the normalized transition probability.

[0054] S90124: Based on the normalized transition probability, construct the migration field matrix: ; Where P represents the migration field matrix, p uv represents the normalized transition probability from state at time u to state at time v, and n represents the total data duration.

[0055] It should be noted that the migration field matrix can extract complex state transition patterns from time series data, not just single acceleration values, but the rules of state transitions in acceleration signals, thereby enhancing the discriminability and robustness of the model.

[0056] S90125: Divide the migration field matrix into multiple sub-blocks, and use the average fuzzy technique to smooth and compress each sub-block, realize the visualization processing of the migration field matrix, and obtain two-dimensional acceleration sensor image data.

[0057] It should be noted that through fuzzy smoothing and compression processing, the high-dimensional transition field matrix can be converted into two-dimensional image data that is easier to process and analyze, which is helpful for subsequent deep learning models to extract features and identify patterns.

[0058] S9013: According to the periodicity of the pumping unit stroke data, the short-time Fourier transform algorithm is used to extract the frequency spectrum of the pumping unit stroke data, and the pumping unit stroke data is converted into two-dimensional pumping unit stroke image data.

[0059] Among them, short-time Fourier transform (STFT) is a method of converting time-domain signals into time-frequency domain images, commonly used to analyze the frequency components of periodic and non-periodic signals. By dividing the signal into several small time windows and performing Fourier transform on each window, STFT can provide the distribution characteristics of the signal in time and frequency. This method can effectively extract the frequency spectrum information of the signal, and is particularly suitable for processing signals with periodic changes, such as pumping unit stroke data. After being converted into two-dimensional images, the frequency spectrum can be used for further pattern recognition and fault detection.

[0060] It should be noted that the pumping unit stroke data presents obvious periodicity, and the short-time Fourier transform is used to extract the frequency spectrum and convert it into a two-dimensional image, so that the periodic characteristics in the original time domain signal can be mapped to the time-frequency domain, the periodic change and frequency distribution are clearly presented, the network can more intuitively identify the abnormal fluctuation of the stroke rule, and the detection capability of the periodic fault is improved.

[0061] In the embodiment of the present application, by selecting a suitable image conversion method according to the characteristics of different sensor data, the dry reed attraction data, the acceleration sensor data and the pumping unit stroke data are uniformly converted into two-dimensional image data, the structural features contained in each type of data can be fully mined, and the data of different modalities are mapped to the same data form, so that the subsequent neural network can be uniformly processed and multi-modal feature fusion, thereby significantly improving the accuracy and robustness of the polished rod rotary actuator operating state monitoring.

[0062] S902: Feature extraction is performed on the two-dimensional image data of each modality to obtain the original feature representation of each modality.

[0063] Specifically, a convolutional neural network can be used to extract features from the two-dimensional image data of each modality to obtain the original feature representation of each modality.

[0064] S903: The original feature representation of each modality is gradually fused for multiple rounds, and in the gradual fusion process, the result of each fusion is projected to an early layer, spliced with the original input and re-input into the feature extraction network, and through multiple rounds of forward fusion, deep-level fusion features are obtained.

[0065] Optionally, S903 specifically includes S9031 to S9035: S9031: In the first round of fusion, the original feature representation of each modality is fused to obtain the fusion feature: ; ; ; Wherein, R1 represents the first round of fusion features, F represents the fusion feature extraction module, X represents the original feature representation, x i represents the original feature representation on the i-th modality, W i represents the weight of the i-th modality, M represents the total number of modalities, and sigma represents the Softmax activation function, W g represents the global weight vector that can be optimized, represents the original feature representation of the k-th sample on the i-th modality, and K represents the total number of samples.

[0066] It should be noted that in the first round of fusion, the original feature representation of each modality is weighted and fused, and then normalized by the Softmax function to participate in fusion. The advantage of this is that the importance of each modality in fusion can be adaptively evaluated and allocated, so that the modality with more abundant information and stronger discriminative ability has a larger weight in fusion, and the weakly related or noisy modality weight is suppressed.

[0067] Optionally, the progressive fusion process (especially the global weight vector) is updated by gradient descent method: ; ; ; wherein, represents the updated global weight vector, L represents the loss function, η represents the learning rate, ∂ represents the partial derivative operation, M represents the total number of modalities, log represents the logarithmic function with base 10, p y represents the probability of the real class y predicted by the model, and λ represents the regularization term coefficient.

[0068] wherein, the gradient descent method is an iterative algorithm for optimizing model parameters, which updates the parameters in the opposite direction of the gradient in each iteration by calculating the gradient of the loss function with respect to the model parameters, so as to minimize the loss function.

[0069] wherein, the skilled person in the art can set the size of the regularization term coefficient according to the actual situation, and the present application does not make any limitation.

[0070] It should be noted that the global weight vector is updated by the gradient descent method, which can dynamically adjust the weight distribution of each modality feature in fusion according to the training error of the model, so that the modality with greater contribution and stronger discriminative ability gradually increases its weight in the training process, while the modality with more noise or weaker correlation is suppressed.

[0071] Optionally, the learning rate is updated based on the decay factor and the gradient adaptive adjustment factor: ; ; ; wherein, η t represents the learning rate at the tth iteration, η0 represents the initial learning rate, s t represents the decay factor at the tth iteration, b t represents the gradient adaptive adjustment factor at the tth iteration, t represents the current iteration number, T max represents the maximum iteration number, exp represents the exponential function with base natural constant, gt denotes the gradient at the tth iteration, g t-1 denotes the gradient at the (t-1)th iteration, a denotes a hyperparameter of the gradient square term, and b denotes a hyperparameter of the gradient change term.

[0072] It should be noted that by setting the learning rate in a form determined by the decay factor and the gradient adaptive adjustment factor, the global convergence speed and the local update stability can be dynamically balanced during the training process. The decay factor gradually decreases with the increase of the iteration number, so that the learning rate is kept larger in the early stage of training to accelerate convergence, and gradually decreases in the later stage to avoid overshooting the optimal solution. The gradient adaptive adjustment factor adaptively scales the learning rate according to the size and change amplitude of the current gradient, automatically reduces the step size when the gradient is large or fluctuates violently, and increases the learning rate when the gradient is small and stable. The advantage of this is that it can accelerate the initial convergence and suppress the later oscillation and overshoot, effectively improving the convergence speed, stability and final accuracy of model training.

[0073] S9032: The fused features are spliced with the original feature representation.

[0074] S9033: The spliced features are down-sampled so that the size of the down-sampled features is the same as that of the original feature representation.

[0075] S9034: The down-sampled features are taken as input, and the next round of fusion is performed: wherein R t+1 denotes the fused features of the t+1th round, R t denotes the fused features of the tth round, Concat denotes the splicing operation, and Downsampling denotes the down-sampling operation.

[0076] S9035: It is determined whether the current round number is greater than the maximum reverse fusion number. If yes, the final deep-level fused features are output. Otherwise, S9032 is returned.

[0077] wherein the skilled person in the art can set the size of the maximum reverse fusion number according to the actual situation, and the present application is not limited.

[0078] ​In the embodiment of the present application, by performing multi-round progressive fusion on the original feature representation of each modality, the result of each round of fusion is projected to the early layer and spliced with the original input to re-input the feature extraction network. The advantage of this is that the fusion of feature information can be deepened layer by layer, so that each round of fusion can further strengthen the relationship and interaction between different modalities, thereby extracting more complex and abstract high-level features. This step-by-step fusion method not only effectively captures the temporal dependence and cross-modal features in different modal data, but also improves the robustness and accuracy of the model, reduces misjudgment caused by insufficient single-modal information, and maintains the scalability and efficiency of the network during each fusion process.

[0079] S904: judging whether the polished rod rotary is normal operation according to the deep fusion feature.

[0080] Optionally, the fusion feature input value is input into a Softmax classifier for state monitoring to judge whether the polished rod rotary is normal operation. The Softmax classifier is a multi-class classification activation function used to convert the original prediction value output by the network into a probability distribution.

[0081] In the embodiment of the present application, according to the dry reed attraction data, acceleration sensor data and pumping unit stroke data, the deep learning algorithm based on the progressive fusion mechanism is used to capture the complex mode in the device operation, accurately judge whether the polished rod rotary is normal operation, and improve the accuracy of the polished rod rotary operation state monitoring.

[0082] The execution subject of the polished rod rotary operation state monitoring method based on multi-source sensor data provided by the embodiment of the present application can be a polished rod rotary operation state monitoring device based on multi-source sensor data. In the embodiment of the present application, the polished rod rotary operation state monitoring device based on multi-source sensor data is taken as an example to illustrate the polished rod rotary operation state monitoring device based on multi-source sensor data provided by the embodiment of the present application.

[0083] Referring to the accompanying drawings Figure 2 , a structure schematic diagram of a polished rod rotary operation state monitoring system based on multi-source sensor data provided by the embodiment of the present application is shown.

[0084] The embodiment of the present application provides a polished rod rotary operation state monitoring system based on multi-source sensor data 20, which comprises a processor 201 and a memory 202. The memory 202 stores programs or instructions executable on the processor 201, which, when executed by the processor 201, implement the steps of the above-mentioned method for monitoring the running state of a polished rod rotary based on multi-source sensor data, and achieve the same technical effects. To avoid repetition, the present application will not repeat the present application will not repeat.

[0085] It should be understood that the processor 201 in the embodiments of the present application can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), ready programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or can also be any conventional processor.

[0086] It should also be understood that the memory 202 in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM) or a flash memory. The volatile memory can be a random access memory (RAM) used as an external cache. By way of example but not limitation, many forms of random access memory are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM).

[0087] The above-described embodiments can be implemented in part or in whole through software, hardware (e.g., circuitry), firmware, or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When loaded and executed by a computer, the computer instructions or computer programs cause the computer to perform the processes or functions described above according to the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, such as from a website, a computer, a server, or a data center to another website, computer, server, or data center through a wired (e.g., infrared, wireless, microwave, or the like) manner. The computer-readable storage medium can be any available medium or a collection of medium accessible by a computer or a data storage device such as a server, data center, or the like, containing one or more of the available medium. The available medium can be a magnetic medium (e.g., a floppy diskette, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0088] It should be understood that the size of the serial number of each process described above does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0089] Those skilled in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0090] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, apparatuses, and units can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0091] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other manners. For example, the embodiments of the apparatus described above are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation. For example, a plurality of units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0092] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0093] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can be physically present alone, or two or more units can be integrated into one unit.

[0094] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the technical solutions that make essential contributions to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0095] The embodiment of the present application provides a readable storage medium, which includes: a program or instruction stored on the readable storage medium, the program or instruction is executed by a processor to realize the steps of the optical rod rotating device running state monitoring method based on multi-source sensor data described above, and the same technical effect can be achieved. To avoid repetition, the present application will not be described again.

[0096] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present application, but not to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered within the protection scope of the present application.

Claims

1. A method for monitoring the operating condition of a polished rod rotary based on multi-source sensor data, characterized in that, Comprise: S1: obtain dry reed pipe attraction data; S2: according to the dry reed pipe attraction data, judge whether the polished rod rotary exists abnormality; If yes, enter next step, activate pumping unit running state monitoring; Otherwise, maintain dormancy, return to S1 and continue to monitor; S3: obtain acceleration sensor data; S4: according to the acceleration variation characteristics of the acceleration sensor data, judge whether the pumping unit is normally operated; If yes, enter next step, activate polished rod rotary running state monitoring; Otherwise, return to S3; S5: according to the acceleration sensor data, calculate pumping unit stroke data; S6: wait for the next dry reed pipe attraction data; S7: according to the pumping unit stroke data and the interval time between two dry reed pipe attractions, calculate the rotation angle of the polished rod rotary; S8: according to the rotation angle of the polished rod rotary, judge whether the polished rod rotary is abnormal; If yes, output the polished rod rotary fault alarm.

2. The light rod rotator operating condition monitoring method based on multi-source sensor data according to claim 1, characterized in that, The S2 specifically is: According to the comparison result of the dry reed pipe attraction time length in the dry reed pipe attraction data and the preset time length, judge whether the polished rod rotary exists abnormality; If the dry reed pipe attraction time length in the dry reed pipe attraction data is greater than the preset time length, enter next step, activate pumping unit running state monitoring; Otherwise, maintain dormancy, return to S1 and continue to monitor.

3. The light rod rotator operating condition monitoring method based on multi-source sensor data according to claim 1, characterized in that, The S4 specifically is: According to the comparison result of the peak-peak value of the acceleration sensor data and the preset peak-peak value, judge whether the pumping unit is normally operated; If the peak-peak value of the acceleration sensor data is greater than the preset peak-peak value, enter next step, activate polished rod rotary running state monitoring; Otherwise, return to S3, the peak-peak value refers to the difference between the maximum value and the minimum value.

4. The light rod rotator operating condition monitoring method based on multi-source sensor data according to claim 1, characterized in that, The S7 specifically is: According to the pumping unit stroke data and the interval time between two dry reed pipe attractions, calculate the rotation angle of the polished rod rotary: Wherein, θ represents the rotation angle of the polished rod rotary, T represents the interval time between dry reed pipe attractions, SPM represents the pumping unit stroke data.

5. The light rod rotator operating condition monitoring method based on multi-source sensor data according to claim 1, characterized in that, The S8 specifically is: According to the comparison result of the rotation angle of the polished rod rotary and the preset rotation angle, judge whether the polished rod rotary is abnormal; If the rotation angle of the polished rod rotary is less than the preset rotation angle, output the polished rod rotary fault alarm.

6. The light rod rotator operating condition monitoring method based on multi-source sensor data according to claim 1, characterized in that, Also include: S9: when the polished rod rotary does not occur abnormality, activate depth state monitoring, according to the dry reed pipe attraction data, the acceleration sensor data and the pumping unit stroke data, judge whether the polished rod rotary is normally operated through the deep learning algorithm based on the progressive fusion mechanism; If yes, return to S1 and continue to monitor; Otherwise, output the polished rod rotary fault alarm; The deep learning algorithm based on the progressive fusion mechanism is used to judge whether the polished rod rotary is normally operated, specifically including: S901: according to the data characteristics of the dry reed pipe attraction data, the acceleration sensor data and the pumping unit stroke data, select the corresponding conversion mode, and convert the dry reed pipe attraction data, the acceleration sensor data and the pumping unit stroke data into two-dimensional image data; S902: feature extraction is performed on the two-dimensional image data of each modality to obtain original feature representations of each modality; S903: multi-round progressive fusion is performed on the original feature representations of each modality, and in the process of progressive fusion, the result of each fusion is projected to an early layer, spliced with the original input, and then input into the feature extraction network for re-extraction, and through multi-round forward fusion, deep-level fusion features are obtained; S904: whether the polished rod rotary device is in normal operation is determined according to the deep-level fusion features.

7. The light rod rotator operating condition monitoring method based on multi-source sensor data according to claim 6, characterized in that, The S901 specifically includes: S9011: according to the data characteristics that the dry reed valve attraction data only has two states of attraction and non-attraction, a binary image conversion method is used to convert the dry reed valve attraction data into two-dimensional dry reed valve attraction image data; S9012: according to the data characteristics that the acceleration sensor data has time sequence dependence, a Markov transition field image conversion algorithm is used to convert the acceleration sensor data into two-dimensional acceleration sensor image data; S9013: according to the data characteristics that the pumping unit stroke data has periodic repetition, a short-time Fourier transform algorithm is used to extract the frequency spectrum of the pumping unit stroke data, and the pumping unit stroke data is converted into two-dimensional pumping unit stroke image data.

8. The light rod rotator operating condition monitoring method based on multi-source sensor data according to claim 7, characterized in that, The S9012 specifically includes: S90121: the one-dimensional acceleration sensor data is binned according to quantiles to obtain multiple states; S90122: transition probabilities between all states are calculated through maximum likelihood estimation; S90123: the transition probabilities are normalized to obtain normalized transition probabilities; S90124: a transition field matrix is constructed based on the normalized transition probabilities; S90125: the transition field matrix is divided into multiple sub-blocks, and each sub-block is smoothed and compressed using an average blur technique to realize visualization processing of the transition field matrix, and two-dimensional acceleration sensor image data is obtained.

9. The light rod rotator operating condition monitoring method based on multi-source sensor data according to claim 6, characterized in that, The S903 specifically includes: S9031: in the first round of fusion, the original feature representations of each modality are fused to obtain fusion features; S9032: the fusion features are spliced with the original feature representations; S9033: the spliced features are down-sampled so that the size of the down-sampled features is the same as that of the original feature representations; S9034: the down-sampled features are input for the next round of fusion; S9035: it is determined whether the current round number is greater than the maximum reverse fusion number; if yes, the final deep-level fusion features are output; otherwise, S9032 is returned.

10. A light rod rotator operating condition monitoring system based on multi-source sensor data, characterized in that, It includes: a processor and a memory; The memory stores programs or instructions that can be run on the processor, and the programs or instructions are executed by the processor to realize the steps of the polished rod rotary device operation state monitoring method based on multi-source sensor data as claimed in any one of claims 1 to 9.

Images