Beam-pumping unit motion parameter estimation method based on multi-scale convolution noise reduction auto-encoder

Abstract

The invention belongs to the technical field of signal processing, and particularly relates to a beam-pumping unit motion parameter estimation method based on a multi-scale convolution noise reduction auto-encoder. Comprising the following steps that acceleration signal data in the operation process of the beam-pumping unit are collected through an acceleration sensor; performing sliding sampling on the processed acceleration data through a sliding rectangular frame with a fixed length to obtain acceleration sampling data with a fixed length; constructing a multi-scale convolution noise reduction auto-encoder model; performing random corrosion processing on the acceleration sampling data to obtain corrosion data; using the sampling data and the corresponding corrosion data to train a multi-scale convolution noise reduction auto-encoder model; acceleration data is collected and processed in real time, and noise reduction processing is carried out on the acceleration data through the trained multi-scale convolution noise reduction auto-encoder model. And calculating motion parameters of the oil pumping unit by using the noise reduction acceleration data, wherein the motion parameters comprise the stroke and the stroke frequency. The method reduces stroke and stroke frequency calculation errors caused by data acquisition loss and noise errors.

Description

Technical Field

[0001] The invention belongs to the technical field of signal processing, in particular to a method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder. Background Art

[0002] The beam pumping unit is an important equipment for oilfield production. Accurate detection and management of it is an important guarantee for the safety and stable production of the oilfield. One of the main means of supervision of the beam pumping unit is to analyze and interpret the indicator diagram. The indicator diagram curve shows the relationship between the pumping unit's donkey head stroke and the load on the donkey head. Because the oilfield production is affected by many factors, such as sand, wax, gas, viscosity, and unknown and uncertain noise, the indicator diagram curve changes complexly. According to the dynamic and static data of the oil well and the pumping unit equipment information, as well as the dynamic physical process analysis of the pumping unit, combined with the indicator diagram curve, the working status of the deep well pump can be inferred. Timely and accurate monitoring of the working status of the deep well pump, timely troubleshooting, and ensuring the stable operation of the pump are important tasks in oilfield management.

[0003] In order to accurately and in real time draw the pumping unit indicator diagram, it is necessary to measure the position of the donkey head stroke. At present, there are three ways to measure the donkey head stroke in oil fields: angular displacement, dead point switch and acceleration sensor measurement. Because the angular displacement measurement method is susceptible to noise interference and has poor stability, and the dead point switch can only measure the moment when the donkey head reaches the upper and lower dead points, and cannot depict the entire stroke of the donkey head, the present invention adopts an acceleration sensor measurement method. The acceleration sensor measurement method collects the acceleration at the suspension point by mounting an acceleration sensor at the suspension point of the walking beam pump, and then integrates the acceleration data twice over time to obtain the suspension point displacement, and then maps it to the donkey head stroke. Most of the data collected by the sensor contains noise, and data missing phenomena will also occur during the collection process. After filtering the collected data by traditional low-pass filtering and other methods, filtering errors will be introduced, and finally the error of the result calculated by the algorithm will also be very large. In view of the influence of noise and data missing in the collected signals of the acceleration sensor measurement method, the present invention adopts a multi-scale convolutional denoising autoencoder to denoise the collected acceleration signal data, and finally calculates the motion parameters of the denoised data: stroke and stroke frequency, thereby greatly improving the accuracy of the calculated motion parameters. Summary of the invention

[0004] In order to solve the problem of accurate estimation of the motion parameters of a beam pumping unit, the present invention aims to provide training and deployment of a multi-scale convolutional denoising autoencoder model, thereby achieving real-time and accurate denoising of the acquired acceleration data, making the stroke and frequency calculated based on the acceleration data more accurate, and reducing the calculation errors of the stroke and frequency caused by data acquisition loss and noise errors.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0006] The motion parameter estimation method of a beam pumping unit based on a multi-scale convolutional denoising autoencoder comprises the following steps:

[0007] 1) The acceleration signal data of the beam pumping unit during operation is collected through the acceleration sensor, and the acceleration data is processed through the WIA-PA chip;

[0008] 2) Sliding sampling is performed on the processed acceleration data through a sliding rectangular frame of fixed length to obtain acceleration sampling data of fixed length;

[0009] 3) Construct a multi-scale convolutional denoising autoencoder model;

[0010] 4) Perform random corrosion processing on the acceleration sampling data to obtain corrosion data;

[0011] 5) Use the sampled data and the corresponding eroded data to train the multi-scale convolutional denoising autoencoder model;

[0012] 6) Using steps 1) and 2), the acceleration data is collected and processed in real time, and denoised using the trained multi-scale convolutional denoising autoencoder model;

[0013] 7) Use the noise-reduced acceleration data to calculate the pumping unit motion parameters: stroke and number of strokes.

[0014] The step 2) comprises the following steps:

[0015] 2.1) Set the sliding rectangle parameters: rectangle width, i.e. fixed time length L, sliding time step P;

[0016] 2.2) Sampling the acceleration data by sliding the rectangular frame to obtain n groups of sampled data D, where the nth group of sampled data D n for:

[0017]

[0018] Among them, A t It represents the acceleration data collected at time t, t 0 is the starting time of acceleration sampling.

[0019] The multi-scale convolutional denoising autoencoder model in the step 3) is composed of an encoder and a decoder, wherein the encoder is composed of multiple layers of multi-scale convolutional layers, and the decoder is composed of multiple layers of multi-scale deconvolutional layers. The encoder and decoder have the same number of layers, and each layer of the decoder is symmetrical with each layer of the encoder.

[0020] The multi-scale convolution layer is composed of one-dimensional convolution kernels with convolution kernels of 3×1, 5×1, 7×1, and 9×1 stacked in sequence, and the multi-scale deconvolution layer is composed of one-dimensional deconvolution kernels with convolution kernels of 3×1, 5×1, 7×1, and 9×1 stacked in sequence. During forward calculation, the feature data are respectively passed through the convolution kernels to obtain the corresponding features, and then the feature data are spliced ​​in the channel dimension to obtain the feature data passing through the multi-scale convolution layer.

[0021] The step 4) comprises the following steps:

[0022] 4.1) Select the corrosion degree C;

[0023] 4.2) Randomly select C×100% of the data points in the sampled data, that is, the acceleration values, and set them to 0, completing the data corrosion process with a corrosion degree of C.

[0024] The step 5) comprises the following steps:

[0025] 5.1) Corrosion data through multi-scale convolutional denoising autoencoder model Perform inference denoising to obtain denoised data results

[0026] 5.2) Calculate the noise reduction data The corresponding original sampling data D n The average of the squared Euclidean distances is used as the loss function l:

[0027]

[0028] 5.3) Update the model parameters θ according to the loss function l:

[0029]

[0030] Among them, γ is the learning rate, is the model parameter gradient;

[0031] 5.4) Repeat steps 5.1) to 5.3) until the loss function is reduced to meet the threshold and no longer decreases, completing the model training.

[0032] The step 7) comprises the following steps:

[0033] 7.1) Denoised acceleration data E n Perform integration and averaging to obtain the time t from the start of deployment monitoring 0 The acceleration data from the beginning to the time t

[0034] 7.2) From the time t when the deployment monitoring starts 0 First, the acceleration data Integrate the time once to obtain the velocity data V corresponding to the time t t

[0035]

[0036] Wherein, Δt represents the time interval between adjacent data points of acceleration data, that is, the time interval for the device to sample acceleration signals;

[0037] 7.3) For speed data V t Integrate the time once to get the donkey head displacement data S at time t t

[0038]

[0039] 7.4) Calculate displacement data S t The maximum and minimum values ​​in , the difference between the adjacent maximum and minimum values ​​is the stroke s at the current moment,

[0040] s=last(maximum(S t ))-last(minimum(S t ))

[0041] Among them, maximum(S t ), minimum(S t ) represent the maximum value and minimum value respectively, and last(*) represents the last value in the * sequence, that is, the value closest to the current moment;

[0042] 7.5) Calculate the stroke f, that is, the pump completes f strokes per unit time. The time T to complete a complete stroke is:

[0043]

[0044] The impulse frequency f is:

[0045] f = 1 / T;

[0046] 7.6) The calculated stroke and stroke frequency are accumulated as the monitoring time continues, and the stroke average value is calculated respectively. Average value of the number of strokes As the final stroke and stroke frequency estimation, to reduce the instantaneous local error.

[0047] The integrated average processing method in step 7.1) is specifically as follows:

[0048]

[0049] in, N is a natural number, upround(*) means rounding up, downround(*) means rounding down, m is the number of natural numbers in the set N, is the data corresponding to time t in the nth sampling data.

[0050] The motion parameter estimation system of the beam pumping unit based on multi-scale convolutional denoising autoencoder includes:

[0051] The data acquisition and processing module is used to collect acceleration signal data during the operation of the beam pumping unit through an acceleration sensor, process the acceleration data through a WIA-PA chip, and perform sliding sampling on the processed acceleration data through a sliding rectangular frame of a fixed length to obtain acceleration sampling data of a fixed length;

[0052] A random corrosion processing module is used to perform random corrosion processing on the acceleration sampling data to obtain corrosion data;

[0053] Model building and training module, used to build a multi-scale convolutional denoising autoencoder model and train the multi-scale convolutional denoising autoencoder model using sampling data and corresponding erosion data;

[0054] Model denoising module, which is used to collect and process acceleration data in real time and denoise it through the trained multi-scale convolutional denoising autoencoder model;

[0055] The pumping unit motion parameter calculation module is used to calculate the pumping unit motion parameters: stroke and stroke frequency using noise reduction acceleration data.

[0056] The present invention has the following beneficial effects and advantages:

[0057] 1. The acceleration denoising technology of the beam pumping unit based on the multi-scale convolutional denoising autoencoder of the present invention has a simple algorithm and can realize fast data preprocessing and denoising within the time length of the sliding rectangular frame. When deployed on the spot, the acceleration denoising data of the corresponding moment can be given in time after each data sampling of the time length of the sliding rectangular frame is completed, and smoother and more accurate denoised acceleration data can be obtained by integrating the average processing method.

[0058] 2. The present invention relates to a method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder. The multi-scale convolutional layer includes multiple convolution kernel scales, which can have more receptive fields when reasoning about data features, so that the reasoning results lose less information about the relationship between features, thereby improving the denoising credibility of the denoising autoencoder.

[0059] 3. After being deployed in the field, the motion parameter estimation method of the walking beam pump based on the multi-scale convolutional denoising autoencoder of the present invention can, as the sampling time increases, on the one hand, obtain the stroke and stroke frequency estimation at the current moment when completing the data sampling of a sliding rectangular frame time length, and can also reduce the error caused by the disturbance in the local time by continuously accumulating the stroke and stroke frequency data results and taking the average value, so as to more accurately estimate the stroke and stroke frequency estimation within the current sampling working time. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of the method of the present invention;

[0061] Figure 2 This is the structure diagram of the multi-scale convolutional layer denoising autoencoder;

[0062] Figure 3 It is a structural diagram of multi-scale convolutional layer and deconvolution layer;

[0063] Figure 4 This is a time series diagram of the acceleration acquisition data and noise reduction data for well 37-75;

[0064] Figure 5 This is a comparison chart of the stroke frequency measured in the field, the stroke frequency calculated by the original signal, and the stroke frequency calculated by this patent. DETAILED DESCRIPTION

[0065] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. According to the following description, the advantages and features of the present invention will be clearer. It should be noted that the drawings are in a very simplified form and use non-precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In order to make the purposes, features and advantages of the present invention more obvious and easy to understand, please refer to the accompanying drawings. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention, so they have no technical substantive significance. Any structural modification, change in proportional relationship or adjustment of size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope of the technical content disclosed by the present invention.

[0066] like Figure 1 The flowchart of the method of the present invention provides a method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder, which specifically includes the following steps:

[0067] Step 1: The acceleration sensor collects the acceleration signal at the suspension point of the beam pumping unit;

[0068] Step 1.1 The acceleration sensor is installed at the suspension point of the beam pumping unit and the signal is collected;

[0069] Step 1.2: Processing the acceleration signal through the WIA-PA chip to obtain the processed acceleration signal data A;

[0070] Step 2 performs sliding sampling on the processed acceleration data through a sliding rectangular frame of fixed length to obtain acceleration sampling data of fixed length, which specifically includes the following steps:

[0071] Step 2.1 Set the sliding rectangle parameters: rectangle width, i.e. fixed time length L = 20s; sliding time step P = 0.5s;

[0072] Step 2.2: The acceleration data A is sampled through the sliding rectangle set in step 2.1, and the nth group of sampled data D is obtained by sampling. n for:

[0073]

[0074] Among them, A t It represents the acceleration data collected at time t, and the starting time of acceleration sampling is t 0 , divide the time into rectangular boxes according to the fixed time length L, and slide backward according to the fixed time interval P, so after sliding n times, the sampled data D is obtained n From time t 0 +n×P to time t 0 +n×P+L acceleration data in the time period;

[0075] Step 2.3: As the acceleration sensor in step 1 continuously collects signals, the acceleration data increases over time. Every time the sampling time increases by one sliding time step, a set of the latest fixed-time sampling data is obtained through the sampling data obtained in step 2.2.

[0076] Step 3 constructs a multi-scale convolutional denoising autoencoder model. The process of constructing the multi-scale convolutional denoising autoencoder model is as follows:

[0077] Step 3.1 constructs multi-scale convolutional layers and deconvolutional layers, such as Figure 3 As shown in the figure, the multi-scale convolution layer is specifically composed of one-dimensional convolution kernels of 3×1, 5×1, 7×1, and 9×1, and the multi-scale deconvolution layer is specifically composed of one-dimensional deconvolution kernels of 3×1, 5×1, 7×1, and 9×1. During forward calculation, the feature data are respectively passed through the convolution kernels to obtain the corresponding features, and then the feature data are spliced ​​in the channel dimension to obtain the feature data passing through the multi-scale convolution layer;

[0078] Step 3.2 The multi-scale convolutional denoising autoencoder consists of two parts: the encoder and the decoder, as shown in Figure 2 As shown, first construct an encoder, the encoder is composed of multiple layers of multi-scale convolutional layers, specifically, in this embodiment, it is composed of three layers of multi-scale convolutional layers, and the number of channels is 8 (wherein the number of channels corresponding to the 3×1, 5×1, 7×1, and 9×1 convolutional layers is 2 each, and the number of channels after splicing is 8), 16 (wherein the number of channels corresponding to the 3×1, 5×1, 7×1, and 9×1 convolutional layers is 4 each, and the number of channels after splicing is 16), and 32 (wherein the number of channels corresponding to the 3×1, 5×1, 7×1, and 9×1 convolutional layers is 8 each, and the number of channels after splicing is 32);

[0079] Step 3.3 Each layer of the decoder is symmetrical with each layer of the encoder, so specifically in this embodiment, the decoder is also composed of three layers of multi-scale deconvolution layers, and the number of channels is 32, 16, and 8 respectively, corresponding to the encoder;

[0080] Step 3.4 After the sampled data passes through the encoder, sampled feature data is obtained, and then the sampled feature data passes through the decoder to obtain noise reduction data;

[0081] Step 4 performs random corrosion processing on the acceleration sampling data to obtain corrosion data. The specific process is as follows:

[0082] Step 4.1: For the sample data D obtained in step 2 n Perform random corrosion treatment, the random corrosion treatment process is specifically as follows:

[0083] Step 4.1.1: First, select the corrosion degree C. Specifically, in this embodiment, select C=0.3;

[0084] Step 4.1.2 Then randomly select C×100% of the data points in the sampled data, that is, the acceleration values, and set them to 0, that is, the data corrosion process with a corrosion degree of C is completed;

[0085] Step 4.2 Because it is a random corrosion process, for the same set of data, each corrosion obtains corrosion data Not completely consistent;

[0086] Step 5 uses the sampled data and the corresponding eroded data to train the multi-scale convolutional denoising autoencoder model. The specific training process is as follows:

[0087] Step 5.1 According to the corrosion process of step 4, the sample data D n After corrosion, the corresponding corrosion data is obtained Using corrosion data The denoised data results are obtained by inference through the multi-scale convolutional denoising autoencoder model.

[0088] Step 5.2 Calculate the denoised data Compared with the original sampling data D n The average of the squared Euclidean distances is used as the loss function l:

[0089]

[0090] Step 5.3 Update the model parameters θ according to the loss function l:

[0091]

[0092] Where γ is the learning rate, which is set to 0.001 in this specific embodiment. is the model parameter gradient;

[0093] Step 5.4: Repeat steps 5.1 to 5.3 until the loss function is reduced to a relatively small value and no longer decreases. The model training is now complete.

[0094] Step 6: Field deployment: Train a multi-scale convolutional denoising autoencoder model, using the sample data D collected in step 2 and processed by the sliding rectangle. n , through the deployed model, reasoning is performed to obtain the noise-reduced acceleration data E n ,like Figure 4 As shown, it is a timing diagram of acceleration acquisition data and noise reduction data of pumping well 37-75;

[0095] Step 7 uses the noise reduction acceleration data to calculate the pumping unit motion parameters: stroke and stroke frequency. The specific process of calculating the pumping unit stroke and stroke frequency is as follows:

[0096] Step 7.1: De-noised acceleration data E obtained in step 6 n Perform integration and averaging to obtain the time t from the start of deployment monitoring 0 The acceleration data from the beginning to the time t The integrated average processing method is specifically as follows:

[0097]

[0098] in, Where N is a natural number, upround(*) means rounding up, downround(*) means rounding down, and m is the number of natural numbers in the set N, which means that there are m sliding rectangles whose sampling data contain acceleration data at time t. Therefore, the average of all acceleration data containing time t is obtained. The data corresponding to time t in the nth sampling data is

[0099] Step 7.2: Deploy monitoring starting at time t0 First, the acceleration data Integrate the time once to obtain the velocity data V corresponding to the time t t ,

[0100]

[0101] Wherein, Δt represents the time interval between adjacent data points of acceleration data, that is, the time interval for the device to sample the acceleration signal. In this patent, Δt=0.1s;

[0102] Step 7.3: Recalculate the speed data V t Integrate the time once to get the donkey head displacement data S at time t t ,

[0103]

[0104] Step 7.4 Calculate displacement data S t The maximum and minimum values ​​in , the difference between the adjacent maximum and minimum values ​​is the stroke s at the current moment,

[0105] s=last(maximum(S t ))-last(minimum(S t ))

[0106] Among them, maximum(S t ), minimum(S t ) represent the maximum value and minimum value respectively, and last(*) represents the last value in the * sequence, that is, the value closest to the current moment;

[0107] Step 7.5 calculates the stroke f, that is, the pump completes f strokes per unit time, and the time T to complete a complete stroke is:

[0108]

[0109] Then the impulse frequency f is:

[0110] f=1 / T

[0111] The stroke and number of strokes calculated in step 7.6 are accumulated as the monitoring time continues, and the stroke average values ​​are calculated respectively. Average value of the number of strokes As the final stroke and stroke frequency estimation, to reduce the instantaneous local error.

[0112] The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder of the present invention realizes the denoising function of acceleration signal data through a multi-scale convolutional denoising autoencoder, and can estimate the stroke and stroke frequency calculated based on the acceleration data, such as Figure 5 As shown in the figure, the field acquisition signals and the stroke and frequency data obtained in some oil wells are closer to the field data than the stroke and frequency results calculated by the original acquisition signals based on the method of denoising the acquisition signals of the present invention, which proves the effectiveness of the method of the present patent. At the same time, the multi-scale convolution kernel enables the denoising autoencoder to have a wider range of receptive fields when performing denoising inference, thereby improving the reliability of denoising. On the other hand, by corroding the training data, the model has better robustness and enhances the ability of denoising.

[0113] The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder of the present invention can perform real-time online data denoising during the deployment phase because the multi-scale convolutional denoising autoencoder occupies less memory and has fewer parameters, thereby greatly improving the real-time monitoring capability of the working condition of the beam pumping unit. The stroke and stroke frequency calculated based on the denoised acceleration data can help staff quickly and accurately grasp the status information of the pumping process, thereby preventing accidents.

[0114] The above description is only an embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, expansion, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder, characterized in that: The following steps are involved: 1) The acceleration signal data of the beam pumping unit during operation is collected through the acceleration sensor, and the acceleration data is processed through the WIA-PA chip; 2) Sliding sampling is performed on the processed acceleration data through a sliding rectangular frame of fixed length to obtain acceleration sampling data of fixed length; 3) Construct a multi-scale convolutional denoising autoencoder model; 4) Perform random corrosion processing on the acceleration sampling data to obtain corrosion data; 5) Use the sampled data and the corresponding eroded data to train the multi-scale convolutional denoising autoencoder model; 6) Using steps 1) and 2), the acceleration data is collected and processed in real time, and denoised using the trained multi-scale convolutional denoising autoencoder model; 7) Use the noise-reduced acceleration data to calculate the pumping unit motion parameters: stroke and number of strokes.

2. The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder according to claim 1 is characterized in that: The step 2) comprises the following steps: 2.1) Set the sliding rectangle parameters: rectangle width, i.e. fixed time length L, sliding time step P; 2.2) Sampling the acceleration data by sliding the rectangular frame to obtain n groups of sampled data D, where the nth group of sampled data D n for: Among them, A t It represents the acceleration data collected at time t, and t0 is the starting time of acceleration sampling.

3. The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder according to claim 1 is characterized in that: The multi-scale convolutional denoising autoencoder model in the step 3) is composed of an encoder and a decoder, wherein the encoder is composed of multiple layers of multi-scale convolutional layers, and the decoder is composed of multiple layers of multi-scale deconvolutional layers. The encoder and decoder have the same number of layers, and each layer of the decoder is symmetrical with each layer of the encoder.

4. The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder according to claim 3 is characterized in that: The multi-scale convolution layer is composed of one-dimensional convolution kernels with convolution kernels of 3×1, 5×1, 7×1, and 9×1 stacked in sequence, and the multi-scale deconvolution layer is composed of one-dimensional deconvolution kernels with convolution kernels of 3×1, 5×1, 7×1, and 9×1 stacked in sequence. During forward calculation, the feature data are respectively passed through the convolution kernels to obtain the corresponding features, and then the feature data are spliced ​​in the channel dimension to obtain the feature data passing through the multi-scale convolution layer.

5. The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder according to claim 1, characterized in that: The step 4) comprises the following steps: 4.1) Select the corrosion degree C; 4.2) Randomly select C×100% of the data points in the sampled data, that is, the values ​​of acceleration, and set them to 0, completing the data corrosion process with a corrosion degree of c.

6. The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder according to claim 1, characterized in that: The step 5) comprises the following steps: 5.1) Corrosion data through multi-scale convolutional denoising autoencoder model Perform inference denoising to obtain denoised data results 5.2) Calculate the noise reduction data The corresponding original sampling data D n The average of the squared Euclidean distances is used as the loss function l: 5.3) Update the model parameters θ according to the loss function l: Among them, γ is the learning rate, is the model parameter gradient; 5.4) Repeat steps 5.1) to 5.3) until the loss function is reduced to meet the threshold and no longer decreases, completing the model training.

7. The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder according to claim 1, characterized in that: The step 7) comprises the following steps: 7.1) Denoised acceleration data E n Perform integration and averaging to obtain the acceleration data from the start time t0 of the deployment monitoring to the time t 7.2) Starting from the deployment monitoring start time t0, the acceleration data Integrate the time once to obtain the velocity data V corresponding to the time t t Wherein, Δt represents the time interval between adjacent data points of acceleration data, that is, the time interval for the device to sample acceleration signals; 7.3) For speed data V t Integrate the time once to get the donkey head displacement data S at time t t 7.4) Calculate displacement data S t The maximum and minimum values ​​in , the difference between the adjacent maximum and minimum values ​​is the stroke s at the current moment, s=last(maximum(S t ))-last(minimum(S t )) Among them, maximum(S t ), minimum(S t ) represent the maximum value and minimum value respectively, and last(*) represents the last value in the * sequence, that is, the value closest to the current moment; 7.5) Calculate the stroke f, that is, the pump completes f strokes per unit time. The time T to complete a complete stroke is: The impulse frequency f is: f = 1 / T; 7.6) The calculated stroke and stroke frequency are accumulated as the monitoring time continues, and the stroke average value is calculated respectively. Average value of the number of strokes As the final stroke and stroke frequency estimation, to reduce the instantaneous local error.

8. The method for estimating motion parameters of a beam pumping unit based on a multi-scale convolutional denoising autoencoder according to claim 7 is characterized in that: The integrated average processing method in step 7.1) is specifically as follows: in, N is a natural number, upround(*) means rounding up, downround(*) means rounding down, m is the number of natural numbers in the set N, is the data corresponding to time t in the nth sampling data.

9. A beam pumping unit motion parameter estimation system based on a multi-scale convolutional denoising autoencoder, characterized in that: include: The data acquisition and processing module is used to collect acceleration signal data during the operation of the beam pumping unit through an acceleration sensor, process the acceleration data through a WIA-PA chip, and perform sliding sampling on the processed acceleration data through a sliding rectangular frame of a fixed length to obtain acceleration sampling data of a fixed length; A random corrosion processing module is used to perform random corrosion processing on the acceleration sampling data to obtain corrosion data; Model building and training module, used to build a multi-scale convolutional denoising autoencoder model and train the multi-scale convolutional denoising autoencoder model using sampling data and corresponding erosion data; Model denoising module, which is used to collect and process acceleration data in real time and denoise it through the trained multi-scale convolutional denoising autoencoder model; The pumping unit motion parameter calculation module is used to calculate the pumping unit motion parameters: stroke and stroke frequency using noise reduction acceleration data.

Images