A water injection pump operation risk identification method based on multi-level judgment

Abstract

The application discloses a kind of based on multilevel judgment's injection pump operation risk identification method, belong to petroleum engineering technical field, technical scheme is, including by exponential weighted moving average and polynomial interpolation method carries out unsupervised risk identification;Supervised risk identification model is constructed using random convolution kernel transformation method extraction time series feature;Combining unsupervised risk identification and supervised risk identification model constructs injection pump operation risk identification model;Injection pump real-time dynamic operation data are input into injection pump operation risk identification model and carry out risk identification, first carry out unsupervised risk identification, when unsupervised risk identification injection pump current operation possibly exists risk, then identify by supervised risk identification model again.The beneficial effects of the application are that: combining unsupervised method and supervised model, obtain the final injection pump operation risk identification model, achieve the effect that unsupervised method preliminarily screens and filters a large number of normal samples, and supervised method improves the accuracy of identification.

Description

Technical Field

[0001] This invention belongs to the field of petroleum engineering technology, specifically relating to a method for identifying operational risks of water injection pumps based on multi-level judgment. Background Technology

[0002] Water-driven oil recovery is a crucial method in current oilfield development. The operational status of the water injection pump system directly impacts the block's fluid supply capacity, thereby affecting well production. Currently, oilfields assess the operational risks of water injection pumps through phased evaluations based on a safety risk matrix. However, operational risks change in real time, and the evolution of accidents is the result of multiple risk factors overlapping. Therefore, real-time risk identification of water injection pumps is necessary.

[0003] In recent years, research on safety risk identification of water injection pumps has been mainly divided into two stages based on the emergence and development of modern detection technologies: The first stage is based on sensor technology for water injection pump fault diagnosis and risk identification. This involves using different signal processing techniques and selecting sensors such as vibration, temperature, and pressure sensors to measure equipment vibration based on different monitoring parameters, thereby achieving state monitoring and risk identification of the water injection pump. The second stage is based on the development of artificial intelligence. Initially, risk identification methods were based on fault mechanisms, studying the fault mechanisms of water injection pumps and the state of the equipment after a fault from a dynamic perspective. Later, identification methods based on signal analysis and processing emerged, using time-domain, frequency-domain, and time-frequency-domain analysis to identify risks in the water injection pump. Currently, traditional risk identification methods for water injection pump stations mainly rely on signal processing technology combined with over-threshold alarms for risk assessment, which leads to frequent alarms and low accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying operational risks of water injection pumps based on multi-level judgment, in order to address the above problems.

[0005] This invention is achieved through the following measures: a method for identifying operational risks of water injection pumps based on multi-level judgment, characterized by comprising:

[0006] S1. Unsupervised risk identification using exponentially weighted moving average and polynomial interpolation methods;

[0007] S2. A supervised risk identification model is constructed by extracting time series features using the random convolution kernel transformation method.

[0008] S3. Construct a risk identification model for water injection pump operation by combining unsupervised risk identification and supervised risk identification models;

[0009] S4. Input the real-time dynamic operation data of the water injection pump into the water injection pump operation risk identification model for risk identification. First, perform unsupervised risk identification. When the unsupervised risk identification indicates that there may be risks in the current operation of the water injection pump, then perform identification through the supervised risk identification model.

[0010] Furthermore, the unsupervised risk identification method using an adaptive threshold-based exponentially weighted moving average and polynomial interpolation includes:

[0011] S11. Determine the adaptive threshold σ;

[0012] S12. Based on the real-time operating data D of the water injection pump, the exponentially weighted moving average method is used for each dimension of the data to predict the value at the current moment, and the predicted value T is obtained. The difference between D and T is calculated and compared with the threshold σ. If it is less than σ, it is determined that the water injection pump is currently operating stably; if it is greater than σ, it is further determined by the polynomial interpolation method.

[0013] S13. For points that exceed the threshold as determined by the exponentially weighted moving average method, a polynomial interpolation method is used for further judgment. The polynomial interpolation method is used to predict the value at the current time to obtain the predicted value T′. The difference between D and T′ is calculated and compared with the threshold σ. If it is less than σ, it is determined that the water injection pump is currently operating stably; if it is greater than σ, it is considered that there may be risks in the current operation of the water injection pump, and further judgment is required through a supervised risk identification model.

[0014] Further, determining the adaptive threshold σ includes:

[0015] Obtain data from k minutes prior to the time point to be detected as the multidimensional time series data to be identified;

[0016] Based on the real-time operation data D of the water injection pump, the k-minute sequence data before the current time for each dimension is analyzed.

[0017] The smoothed time series S is obtained by using the exponentially weighted moving average method. The difference L between D and S is calculated, and the proportion P of the change in D is calculated.

[0018] For each dimension of data, calculate the average value p of P. p is the average deviation of the real-time data during that period. Set the adaptive threshold σ = 1.5 * p.

[0019] Furthermore, a supervised risk identification model is constructed by extracting time series features using the random convolution kernel transformation method, including:

[0020] S21. Select historical operating data of the water injection pump, including normal operation data and abnormal operation data, to form the training set D. train Each sample consists of three hours of real-time data [D1, D2, ..., D...].i ,...D n It consists of ] and data labels, where D i =[d i1 ,d i2 ,...,d im [] represents the time series corresponding to a certain feature, where m is the length of the time series and n is the total number of features;

[0021] S22, Using n kernel Each random convolution kernel performs convolution calculations on the time series, and for each kernel, the convolutional value C = [c1, c2, c3] is calculated. input -c kernel The proportion of positive values ​​(ppv) and the maximum value (max) are used as time series features for subsequent classification models.

[0022] S23. Using the features and corresponding sample labels obtained in S22 as samples, train a linear classifier to distinguish between normal and abnormal working conditions. Repeatedly train the classifier until the model converges or reaches the preset maximum number of training iterations to obtain the network model.

[0023] Furthermore, in the supervised risk identification model, the size of the Sobel convolution kernel is set to n×3, where n is the number of indicators and 3 represents the window size for feature extraction in each dimension. The values ​​in the first column of the convolution kernel are all set to -1, the values ​​in the middle column are all set to 0, the values ​​in the third column are all set to 1, and the stride is set to 1.

[0024] Furthermore, when using a supervised risk identification model, the obtained window data is first padded with a zero on both the left and right sides to ensure that the horizontal dimension remains unchanged after convolution. After convolution with a kernel, the feature value P at each time point is obtained. i ;

[0025] Take P i The absolute value C is used to determine whether the point is an abnormal alarm point. If C > σ, the point is considered to be under normal operating conditions.

[0026] In S22, the specific settings for selecting the convolution kernel are as follows: number of convolution kernels n kernel =10000, kernel length is taken from 7, 9, 11, weights are sampled from normal distribution N(0,1), bias is sampled from uniform distribution U(-1,1), and inflation coefficient is... Among them x~U(0,A), (l input l is the length of the time series. kernel (where is the kernel length), and padding can be randomized or not padded.

[0027] Furthermore, real-time data from abnormal operating conditions were used as a test set to test the risk identification model for water injection pump operation.

[0028] Furthermore, this embodiment of the invention provides a storage medium, characterized in that the storage medium stores one or more programs, which can be executed by one or more processors to implement the above-described method for identifying the operational risks of a water injection pump based on multi-level judgment.

[0029] Furthermore, this embodiment of the invention provides an electronic device, characterized in that it includes: a processor and a memory, wherein the processor is used to execute a program stored in the memory for identifying the operation risk of a water injection pump based on multi-level judgment, so as to implement the above-described method for identifying the operation risk of a water injection pump based on multi-level judgment.

[0030] The beneficial effects of the technical solution provided by this invention are as follows: It proposes a method for identifying the operational risks of water injection pumps, which involves initial screening using an unsupervised method and further judgment using a feature extraction method. The unsupervised method can initially screen and filter a large number of normal samples, quickly eliminating those operating under normal conditions and improving the method's judgment speed. The feature extraction model further judges abnormal items identified in the unsupervised method, improving accuracy and reducing false alarms. Introducing a dynamic threshold into the unsupervised method allows for better adaptation to different types of water injection pumps. The feature extraction method uses Sobel convolution kernels to achieve deeper feature extraction of time series data, improving the model's judgment accuracy. Attached Figure Description

[0031] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings listed below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Fig. 1 This is a flowchart illustrating the steps of a method for identifying operational risks of a water injection pump based on multi-level judgment in an embodiment of the present invention.

[0033] Fig. 2 This is a schematic diagram of the actual prediction process of a multi-level judgment-based method for identifying operational risks of water injection pumps. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. Of course, the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0035] Example 1:

[0036] See Figs. 1-2 A method for identifying operational risks of water injection pumps based on multi-level judgment, characterized by comprising:

[0037] Step S1: Unsupervised risk identification is performed using exponentially weighted moving average and multinomial interpolation methods, specifically including:

[0038] S11. Determine the adaptive threshold σ; Obtain the data from k minutes before the time point to be detected as the multidimensional time series data to be identified. Based on the real-time operation data D of the water injection pump, apply the exponential weighted moving average method to the k-minute sequence data before the current time of each dimension to obtain the smoothed time series S. Calculate the difference L between D and S, and calculate the proportion P of the change in D. Calculate the average value p of P for each dimension of the data, which is the average deviation of the real-time data during that period. Set the adaptive threshold σ = 1.5 * p.

[0039] S12. Based on the real-time operation data D of the water injection pump, the exponential weighted moving average method is applied to the data of each dimension. Using the sequence data of the previous k minutes before the current time, the value of the current time is predicted to obtain the predicted value S. The difference between D and S is calculated and compared with the threshold σ. If it is less than σ, it is determined that the water injection pump is currently operating stably; if it is greater than or equal to σ, it is further determined by the polynomial interpolation method.

[0040] S13. For points that exceed the threshold as determined by the exponentially weighted moving average method, a polynomial interpolation method is used for further judgment. The polynomial interpolation method is used to predict the value at the current time to obtain the predicted value T′. The difference between D and T′ is calculated and compared with the threshold σ. If it is less than σ, it is determined that the water injection pump is currently operating stably; if it is greater than or equal to σ, it is considered that the current operation of the water injection pump may be at risk and further judgment is required through a supervised risk identification model.

[0041] Step S2: Extract time series features using the random convolution kernel transformation method to construct a supervised risk identification model, specifically including:

[0042] S21. Select historical operating data of the water injection pump (including normal operation and abnormal operation) to form the training set D. train Each sample consists of three hours of real-time data [D1, D2, ..., D i ,…,D n [(where D) i =[d i1 ,d i2 ,…,d im [A] is a time series corresponding to a certain feature, where m is the length of the time series and n is the total number of features) and data labels l and l.

[0043] S22, Using nkernel Each random convolution kernel performs convolution calculations on the time series. The specific settings for selecting the convolution kernels are as follows: Number of convolution kernels n kernel =10000, kernel length is taken from 7, 9, 11, weights are sampled from normal distribution N(0,1), bias is sampled from uniform distribution U(-1,1), and inflation coefficient is... Among them x~U(0,A), (l input l is the length of the time series. kernel (where c is the kernel length), padding can be random or no padding. For each kernel, the convolutional value C = [c1, c2, c3] is calculated. input -c kernel The proportion of positive values ​​(ppv) and the maximum value (max) are used as time series features for subsequent classification models.

[0044] S23. Using the features and corresponding sample labels obtained in step S22 as samples (for each time series, we use 10,000 convolutional kernels for transformation, resulting in 20,000 features), train a linear classifier for judging normal and abnormal operating conditions. Here, Ridge ClassifierCV is selected, with regularization strength set to (-3, 3, 10), and normalization is used during calculation. The above classifier is trained iteratively until the model converges or reaches the preset maximum number of training iterations, resulting in the network model.

[0045] Step S3: Construct a water injection pump operation risk identification model by combining unsupervised and supervised risk identification models. Additionally, collect real-time data on abnormal operating conditions of the sucker pipe that were not used in the training process, and use this data as a test set to test the model. Calculate the model's accuracy, precision, and recall parameters to measure its specific performance.

[0046] Step S4: Input the real-time dynamic operation data of the water injection pump into the water injection pump operation risk identification model for risk identification. First, perform unsupervised risk identification. When the unsupervised risk identification indicates that there may be risks in the current operation of the water injection pump, then perform identification through the supervised risk identification model.

[0047] Example 2:

[0048] See Figs. 1-2 Step S1: Unsupervised risk identification is performed using exponentially weighted moving average and polynomial interpolation methods, specifically including:

[0049] S11. Obtain the data from k minutes prior to the time point to be detected as the multidimensional time series data to be identified, D = [D1, D2, ..., D...]. i ,…,D n ], D i =[di1 ,d i2 ,…,d im Let S be the time series corresponding to a certain feature, i be the i-th dimension of the time series data, m be the length of the time series, and n be the total number of features. For the data k minutes prior to the current time, an exponentially weighted moving average method is used to obtain the smoothed time series S = [S1, S2, ..., Sn]. i ,…,S n ], S i The specific calculation method is as follows:

[0050] S i0 =D i0 ,S it =βS it-1 +(1-β)D it ,t=1,…,m-1

[0051] Calculate time series data D i and smoothed time series S i The difference L i =|D i -S i | and calculate L i Relative to D i The change ratio P i =L i / D i Calculate P i average value P ij This refers to the average change rate of real-time data over that period, with an adaptive threshold σ set. i =1.5*p i .

[0052] S12. Based on the real-time operating data D of the water injection pump, the exponentially weighted moving average method is applied to each dimension of the data to predict the value at the current moment, obtaining the predicted value T. The difference between D and T is calculated and compared with a threshold σ. If it is less than σ, the water injection pump is considered to be operating stably; if it is greater than or equal to σ, a polynomial interpolation method is used for further evaluation. Based on the real-time operating data D of the water injection pump, the exponentially weighted moving average method is applied to each dimension of the data to predict the value at the current moment, obtaining the predicted value S at the final moment m. im Specifically:

[0053] S im =βS im-1 +(1-β)D im

[0054] Calculate D im With S im The difference, and the threshold σ iCompare them. If |D im -S im |<σ i Then it is determined that the water injection pump is currently operating stably; if |D im -T im |≥σ i Then, a second judgment is made using the polynomial interpolation method.

[0055] S13. For points that exceed the threshold as determined by the exponentially weighted moving average method, a polynomial interpolation method is used for further judgment. By fitting a third-order polynomial to the 30 data points before the point to be judged, a cubic function equation is obtained. The polynomial interpolation method is used to predict the value at the current time, and the predicted value S' of the point to be judged is obtained through the equation. im Calculate D im With S' im The difference, and the threshold σ i Compare them. If |D im -S' im |<σ i If |D im -S' im |≥σ i If the current operation of the water injection pump is deemed to pose a risk, a monitoring model is needed for further assessment.

[0056] Step S2: Extract time series features using the random convolution kernel transformation method to construct a supervised risk identification model, specifically including:

[0057] S21. Select historical operating data of the water injection pump (including normal operation and abnormal operation) to form the training set D. train Each sample consists of three hours of real-time data [D1, D2, ..., D i ,…,D n [(where D) i =[d i1 ,d i2 ,…,d im [A time series corresponding to a certain feature, where m is the length of the time series and n is the total number of features] and data labels are used.

[0058] S22, Using n kernel Each random convolution kernel performs convolution calculations on the time series. The specific settings for selecting the convolution kernels are as follows: Number of convolution kernels n kernel =10000, kernel length is taken from 7, 9, 11, weights are sampled from normal distribution N(0,1), bias is sampled from uniform distribution U(-1,1), and inflation coefficient is... Among them x~U(0,A), (l input l is the length of the time series.kernel (where c is the kernel length), padding can be random or no padding. For each kernel, the convolutional value C = [c1, c2, c3] is calculated. input -c kernel The proportion of positive values ​​(ppv) and the maximum value (max) are used as time series features for subsequent classification models.

[0059] S23. Using the features and corresponding sample labels obtained in step S22 as samples (for each time series, we use 10,000 convolutional kernels for transformation, resulting in 20,000 features), train a linear classifier for judging normal and abnormal operating conditions. Here, we choose Ridge ClassifierCV, with regularization strength set to (-3, 3, 10), and normalization is used during calculation. Iteratively train the above classifier until the model converges or reaches the preset maximum number of training iterations, thus obtaining the network model.

[0060] Step S3: Construct a water injection pump operation risk identification model by combining unsupervised and supervised risk identification models. Additionally, collect real-time data on abnormal operating conditions of the sucker pipe that were not used in the training process, and use this data as a test set to test the model. Calculate the model's accuracy, precision, and recall parameters to measure its specific performance.

[0061] Step S4: Input the real-time dynamic operation data of the water injection pump into the water injection pump operation risk identification model for risk identification. First, perform unsupervised risk identification. If the unsupervised risk identification indicates that there may be risks in the current operation of the water injection pump, then perform identification through the supervised risk identification model.

[0062] Example 3:

[0063] See Figs. 1-2 Based on Example 1 or Example 2, 13 real-time dynamic operating data points (e.g., inlet pressure, instantaneous flow rate, etc.) of a certain water injection pump are extracted. Each data point contains 180 time-series data points, the format of which has been given in step S21. In the unsupervised risk identification stage, within a small window of each data point (e.g., the first 60 data points; if the number of data points in the window is less than 60, then 0 is filled in before the data), the predicted value of the last point in the window is calculated using the exponentially weighted moving average method. The difference between the predicted value and the actual result value is calculated. If the difference between the data points at this time point is greater than or equal to the adaptive threshold σ, then this time point is temporarily designated as an abnormal data point.

[0064] If the point is identified as an outlier by the exponentially weighted moving average method, the predicted value of the point is calculated using the polynomial interpolation method within a larger window (window size greater than 60 and less than 180). If the result is less than σ, the water injection pump at that point is considered to be operating normally; otherwise, further judgment is made, and stage B is executed.

[0065] In the supervised risk identification model stage, 13 data points are padded with a 0 on the left and right. The Sobel convolution kernel is used to extract the features of the 13 data points at the same time to obtain the feature value. If the absolute value of the feature value is greater than or equal to σ, the water pump at that point is considered to be operating abnormally; otherwise, the point is considered to be in normal working condition, and the execution ends.

[0066] Example 4:

[0067] This invention provides an electronic device, characterized in that it includes a processor and a memory, wherein the processor is used to execute a program stored in the memory for identifying the operation risk of a water injection pump based on multi-level judgment, so as to implement the above-described method for identifying the operation risk of a water injection pump based on multi-level judgment.

[0068] Furthermore, the electronic device includes a processor, storage units, storage media, and a communication interface. The processor, storage units, storage media, and communication interface communicate via a bus, or through other means such as wireless transmission. The storage unit stores program code and its execution is controlled by the processor. The storage unit includes read-only memory and random access memory (RAM), and provides instructions and data to the processor. The storage unit also includes non-volatile RAM. The storage unit may be volatile memory or non-volatile memory, or a combination of both.

[0069] A bus is a standard bus for rapid interconnection of peripheral components, or an extended industrial standard structure bus, unified bus, computer fast link, cache-consistent interconnection protocol, etc. Buses are divided into address buses, data buses, control buses, etc. In addition to data buses, buses also include power buses, control buses, and status signal buses.

[0070] The processor consists of at least one general-purpose processor, such as a CPU, NPU, or a combination of a CPU and a hardware chip. The aforementioned hardware chip is an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The aforementioned PLD is a complex programmable logic device (PLD), a field-programmable gate array (FPGA), a general-purpose array logic (GPRS), or any combination thereof. The processor executes various types of digital memory instructions, such as software or firmware programs stored in memory cells, enabling the computing device to provide a wide range of services.

[0071] Example 5:

[0072] This invention provides a storage medium, characterized in that the storage medium stores one or more programs, which can be executed by one or more processors to implement the above-described method for identifying the operational risks of water injection pumps based on multi-level judgment.

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying operational risks of water injection pumps based on multi-level judgment, characterized in that, include: S1. Unsupervised risk identification using exponentially weighted moving average and polynomial interpolation methods; S2. A supervised risk identification model is constructed by extracting time series features using the random convolution kernel transformation method. S3. Construct a risk identification model for water injection pump operation by combining unsupervised risk identification and supervised risk identification models; S4. Input the real-time dynamic operation data of the water injection pump into the water injection pump operation risk identification model for risk identification. First, perform unsupervised risk identification. When the unsupervised risk identification indicates that there may be risks in the current operation of the water injection pump, then perform identification through the supervised risk identification model.

2. The method for identifying operational risks of water injection pumps based on multi-level judgment according to claim 1, characterized in that, The method of unsupervised risk identification using an adaptive threshold-based exponentially weighted moving average and polynomial interpolation includes: Determine the adaptive threshold σ; Based on the real-time operating data D of the water injection pump, the exponentially weighted moving average method is used for each dimension of the data to predict the value at the current moment, and the predicted value T is obtained. The difference between D and T is calculated and compared with the threshold σ. If it is less than σ, it is determined that the water injection pump is currently operating stably; if it is greater than σ, a further judgment is made by using a polynomial interpolation method. For points that exceed the threshold as determined by the exponentially weighted moving average method, a polynomial interpolation method is used for further judgment. The polynomial interpolation method is used to predict the value at the current time to obtain the predicted value T′. The difference between D and T′ is calculated and compared with the threshold σ. If it is less than σ, the water injection pump is judged to be operating stably. If it is greater than σ, the water injection pump is considered to be operating at risk and further judgment is required through a supervised risk identification model.

3. The method for identifying operational risks of water injection pumps based on multi-level judgment according to claim 2, characterized in that, Determining the adaptive threshold σ includes: Obtain data from k minutes prior to the time point to be detected as the multidimensional time series data to be identified; Based on the real-time operation data D of the water injection pump, the k-minute sequence data before the current time for each dimension is analyzed. The smoothed time series S is obtained by using the exponentially weighted moving average method. The difference L between D and S is calculated, and the proportion P of the change in D is calculated. For each dimension of data, calculate the average value p of P. p is the average deviation of the real-time data during that period. Set the adaptive threshold σ = 1.5 * p.

4. The method for identifying operational risks of water injection pumps based on multi-level judgment according to claim 3, characterized in that, A supervised risk identification model is constructed by extracting time series features using the random convolution kernel transformation method, including: S21. Select historical operating data of the water injection pump, including normal operation data and abnormal operation data, to form the training set D. train Each sample consists of three hours of real-time data [D1, D2, ..., D...]. i ,...D n It consists of ] and data labels, where D i =[d i1 ,d i2 ,...,d im [] represents the time series corresponding to a certain feature, where m is the length of the time series and n is the total number of features; S22, Using n kernel Each random convolution kernel performs convolution calculations on the time series, and for each kernel, the convolutional value C = [c1, c2, c3] is calculated. input -c kernel The proportion of positive values ​​(ppv) and the maximum value (max) are used as time series features for subsequent classification models. S23. Using the features and corresponding sample labels obtained in S22 as samples, train a linear classifier to distinguish between normal and abnormal working conditions. Repeatedly train the classifier until the model converges or reaches the preset maximum number of training iterations to obtain the network model.

5. The method for identifying operational risks of water injection pumps based on multi-level judgment according to claim 4, characterized in that, In the supervised risk identification model, the size of the Sobel convolution kernel is set to n×3, where n is the number of indicators and 3 represents the window size for feature extraction in each dimension. The values ​​in the first column of the convolution kernel are all set to -1, the values ​​in the middle column are all set to 0, the values ​​in the third column are all set to 1, and the stride is set to 1.

6. The method for identifying operational risks of water injection pumps based on multi-level judgment according to claim 5, characterized in that, When using a supervised risk identification model, the obtained window data is first padded with a zero on both the left and right sides to ensure that the horizontal dimension remains unchanged after convolution. After convolution with a kernel, the feature value P at each time point is obtained. i ; Take P i The absolute value C is used to determine whether the point is an abnormal alarm point. If C > σ, the point is considered to be under normal operating conditions.

7. The method for identifying operational risks of water injection pumps based on multi-level judgment according to claim 5, characterized in that, In S22, the specific settings for selecting the convolution kernel are as follows: number of convolution kernels n kernel =10000, kernel length is taken from 7, 9, 11, weights are sampled from normal distribution N(0,1), bias is sampled from uniform distribution U(-1,1), and inflation coefficient is... Among them x~U(0,A), l input l is the length of the time series. kernel The length of the convolution kernel is specified, and padding can be set to random padding or no padding.

8. The method for identifying operational risks of water injection pumps based on multi-level judgment according to claim 5, characterized in that, Real-time data from abnormal operating conditions were used as a test set to test the risk identification model for water injection pump operation.

9. A storage medium, characterized in that, The storage medium stores one or more programs, which can be executed by one or more processors to implement the water injection pump operation risk identification method based on multi-level judgment as described in any one of claims 1 to 8.

10. An electronic device, characterized in that, include: A processor and a memory, the processor being configured to execute a program stored in the memory for identifying the operation risk of a water injection pump based on multi-level judgment, to implement the water injection pump operation risk identification method based on multi-level judgment as described in any one of claims 1 to 8.

Images