Pump inspection period prediction method and device for rod-pumped well
By fusing the Transform algorithm with the ResNet algorithm, the global features of the pumping well are extracted and feature mapped, the problems of high difficulty in feature extraction and low prediction accuracy in the prior art are solved, and higher prediction accuracy and accuracy are achieved.
Patent Information
- Application Number
- CN202311256833.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-26
- Publication Date
- 2025-03-28
AI Technical Summary
The existing oil pump cycle prediction method for well inspection pumps in the oil pump is difficult to extract features and the accuracy of prediction results is low.
The fusion of Transform algorithm and ResNet algorithm is adopted to extract global features through Transform and use ResNet for feature mapping to generate a pump cycle prediction model based on Transform-ResNet.
The accuracy of the prediction model of the well inspection pump cycle of the oil pump is improved and the accuracy and effectiveness of the prediction results are achieved.
Smart Images
Figure CN119719727A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of oil production engineering, and in particular to a method and a device for predicting a pump inspection cycle of an oil well. Background Art
[0002] Rod pumping units have been widely used in oil production, and pumping well failure has become one of the problems that seriously affect oil field production. Effective prediction of pumping well inspection cycle is of great significance to improving the production efficiency of oil wells. Researchers have done a lot of research work in the diagnosis of pumping well faults and prediction of pumping cycle. At present, the prediction research on pumping well inspection cycle is mainly carried out from two aspects: on the one hand, it is based on the method of mechanism analysis, but the factors affecting the failure of pumping wells are extremely complex and difficult to quantify. The parameters that reflect the working state of the oil well have a weak correlation with the inspection cycle and the parameters affect each other. The accuracy of the method based on mechanism analysis is low; on the other hand, the method is based on data perspective, and the relationship between the production data of the pumping well and the inspection cycle is mined based on data analysis methods. In recent years, the research methods of rod pumping well fault diagnosis technology mainly include expert systems, computer diagnosis, support vector machines and artificial neural networks.
[0003] Among the existing methods, there is a study on the failure rate of oilfield pumping wells based on Weibull distribution, using simulated annealing (FPSO-SA) algorithm to estimate the parameters of the distribution model, establishing the failure rate function of pumping wells, and using fuzzy decision analysis to calculate the optimal preventive pump inspection cycle; and a fault diagnosis method based on gray matrix extreme learning machine (GM-ELM), using gray matrix to extract fault features of rod pumping wells, establishing gray matrix feature vectors, using fault feature vectors as input values of fault diagnosis model, and finally establishing a diagnosis model for rod pumping wells. However, there are many production characteristic parameters of pumping wells, complex factors causing failures, and great difficulty in feature extraction. It is necessary to combine the advantages of multiple algorithms, further carry out data analysis and modeling research, and improve the effectiveness and accuracy of the prediction model. Summary of the invention
[0004] The present invention proposes a method and device for predicting the pump inspection cycle of a pumping well, so as to solve the problem that the existing method for predicting the pump inspection cycle of a pumping well needs to extract key parameters and fault characteristics of the pumping well, but because there are too many characteristic parameters of faults caused by the production of the pumping well, it is difficult to extract the characteristics, resulting in low accuracy of the final prediction result.
[0005] According to one aspect of the present invention, a method for predicting a pump inspection cycle of a well pumping unit is provided, comprising:
[0006] Obtain existing production data of pump inspection cycles of oil wells and establish a sample data set;
[0007] Performing normalization preprocessing on all production data in the sample data set, and dividing the normalized sample data set into a training set and a test set;
[0008] Establishing a Transform-based feature extraction module, the input of the feature extraction module is the preprocessed sample data set, and using the feature extraction module to extract feature vectors in the sample data set;
[0009] Establishing a feature mapping module based on ResNet, wherein the input of the feature mapping module is the feature vector, and using the feature mapping module to generate a pump inspection cycle prediction model for a pumping well based on Transform-ResNet;
[0010] The pump inspection cycle prediction model is trained and tested according to the training set and the test set, and the pump inspection cycle of the target pumping well is predicted using the trained and tested pump inspection cycle prediction model for the pumping well.
[0011] Preferably, the existing production data of the pump inspection cycle of the oil well pumping unit is:
[0012] The pump inspection cycle production data of the pumping wells that are in the same block as the target block or have the same reservoir geological conditions as the target block and the same production conditions.
[0013] Preferably, the production data at least includes:
[0014] Daily liquid production, daily oil production, upstream current, downstream current, oil pressure, casing pressure, water content, stroke, number of strokes, pump depth and pump diameter.
[0015] Preferably, the normalization formula includes:
[0016]
[0017] In the formula, is the normalized production data, x is the original production data of the pumping well, X min is the minimum value in the production data, X max This is the maximum value in the production data.
[0018] Preferably, before the normalization process, the method further comprises:
[0019] If there are gaps in the obtained production data of the pump inspection cycle of the existing pumping well, the missing data are filled by taking the average value of the predetermined number of data before and after the missing data.
[0020] Preferably, the Transform-based feature extraction module comprises a multi-head self-attention layer and a feed-forward neural network layer, wherein the operation process of the self-attention layer includes:
[0021] Embed the data in the preprocessed sample data set into the multi-head self-attention layer in the feature extraction module to generate a query vector, a key vector and a value vector respectively;
[0022] Determine an attention score using a Softmax function according to the query vector and the key vector;
[0023] Determining an attention feature map according to the attention score and the value vector;
[0024] The attention feature maps are combined to obtain multi-head self-attention results.
[0025] Preferably, the ResNet-based feature mapping module includes a residual module, a fully connected layer and a loss function, and the residual module includes a convolutional layer and an activation function;
[0026] Wherein, the loss function is:
[0027]
[0028] Where P j is the predicted pump inspection cycle value output by the model, Y j It is the actual pump inspection cycle value.
[0029] According to one aspect of the present invention, a device for predicting a pump inspection cycle of a well pumping unit is provided, comprising:
[0030] An acquisition unit, used to acquire existing production data of the pump inspection cycle of the pumping well and establish a sample data set;
[0031] A preprocessing unit, used to perform normalization preprocessing on all production data in the sample data set, and divide the normalized sample data set into a training set and a test set;
[0032] A feature extraction unit, used to establish a Transform-based feature extraction module, the input of the feature extraction module is the preprocessed sample data set, and the feature extraction module is used to extract feature vectors in the sample data set;
[0033] A model generation unit, used to establish a feature mapping module based on ResNet, the input of the feature mapping module is the feature vector, and the feature mapping module is used to generate a pump inspection cycle prediction model for a pumping well based on Transform-ResNet;
[0034] The prediction unit trains and tests the pump inspection cycle prediction model according to the training set and the test set, and predicts the pump inspection cycle of the target pumping well by using the trained and tested pump inspection cycle prediction model.
[0035] The present invention has at least the following beneficial effects:
[0036] The present invention proposes a method and device for predicting the pump inspection cycle of a pumping well. By fusing the Transform algorithm with the Resnet algorithm, relying on the Transform algorithm to extract global features and the Resnet algorithm to perform feature mapping, a pump inspection cycle prediction model for a pumping well based on Transform-ResNet is generated, which greatly improves the accuracy of the pump inspection cycle prediction model for the pumping well as the precision and effectiveness of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] The accompanying drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present invention and, together with the specification, are used to explain the technical solutions of the present invention.
[0038] Figure 1 A flow chart of a method for predicting a pump inspection cycle of a well pumping unit according to an embodiment of the present invention is shown. DETAILED DESCRIPTION
[0039] Various exemplary embodiments, features and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.
[0040] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.
[0041] The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set consisting of A, B, and C.
[0042] In addition, in order to better illustrate the present invention, numerous specific details are provided in the following specific embodiments. It should be understood by those skilled in the art that the present invention can be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present invention.
[0043] Figure 1 FIG. 2 is a flow chart showing a method for predicting a pump inspection cycle of a well pumping unit according to an embodiment of the present invention. Figure 1 As shown, a method for predicting the pump inspection cycle of a pumping well comprises: step S01: obtaining existing production data of the pump inspection cycle of the pumping well and establishing a sample data set; step S02: performing normalization preprocessing on all production data in the sample data set, and dividing the normalized sample data set into a training set and a test set; step S03: establishing a feature extraction module based on Transform, the input of the feature extraction module is the preprocessed sample data set, and the feature extraction module is used to extract feature vectors in the sample data set; step S04: establishing a feature mapping module based on ResNet, the input of the feature mapping module is the feature vector, and the feature mapping module is used to generate a pump inspection cycle prediction model for the pumping well based on Transform-ResNet; step S05: training and testing the pump inspection cycle prediction model according to the training set and the test set, and predicting the pump inspection cycle of the target pumping well using the pump inspection cycle prediction model for the pumping well after training and testing.
[0044] A method for predicting a pump inspection cycle of a well pumping unit provided by an embodiment of the present invention specifically comprises the following steps:
[0045] Step S01: Acquire existing production data of the pump inspection cycle of the oil pumping well and establish a sample data set.
[0046] In the present invention, the existing pump inspection cycle production data of the pumping wells are: the pump inspection cycle production data of the pumping wells which are in the same block as the target block or have the same reservoir geological conditions as the target block and have the same production conditions.
[0047] In the present invention, the production data at least includes: daily liquid production, daily oil production, upstream current, downstream current, oil pressure, casing pressure, water content, stroke, stroke times, pump depth and pump diameter.
[0048] In the embodiment of the present invention, the existing pump inspection cycle production data can collect existing production data of other wells in the same block, or other blocks, but the target wells in the target block need to have the same reservoir geological conditions and production conditions.
[0049] The production data of the existing pumping wells during the pump inspection cycle are obtained to form a sample data set; the sample data set means that the complete production data of a pump inspection cycle is one sample data; the production data of the pumping wells contained in the sample data set should at least include: daily liquid production, daily oil production, upstream current, downstream current, oil pressure, casing pressure, water content, stroke, stroke frequency, pump depth, pump diameter, and other production-related indicator data; by obtaining comprehensive production data to participate in subsequent feature extraction and model building, the final model can be made more accurate, and the subsequent pump inspection cycle prediction results can be more realistic and reliable.
[0050] Step S02: performing normalization preprocessing on all production data in the sample data set, and dividing the normalized sample data set into a training set and a test set.
[0051] In the present invention, the formula for the normalization process includes:
[0052]
[0053] In the formula, is the normalized production data, x is the original production data of the pumping well, X min is the minimum value in the production data, X max This is the maximum value in the production data.
[0054] In the present invention, before the normalization process, the method further includes: if there are missing data in the obtained existing pumping well pump inspection cycle production data, the missing data is filled by taking the average value of a predetermined number of data before and after the missing data.
[0055] In the embodiment of the present invention, the predetermined number is 30. When filling in the missing data in the acquired production data, the missing data is filled in by taking the average value of the first 30 and the last 30 data, thereby ensuring the completeness and sufficiency of the original data and the accuracy and credibility of the prediction results.
[0056] Before training the model, the production data in the sample data set needs to be normalized and all production data are scaled between [0,1] according to specific rules. Normalization makes incomparable data comparable while maintaining the relative relationship between the two compared data, reducing the complexity of the calculation process and solving the problem that different production data have different dimensions and affect the training results.
[0057] The normalized sample data set is divided into a training set and a test set, with the proportions of 70% and 30% respectively. The training set and the test set are used for the subsequent training and testing of the pump inspection cycle prediction model for pumping wells based on the Transform-ResNet (residual neural network) algorithm.
[0058] Step S03: establishing a Transform-based feature extraction module, wherein the input of the feature extraction module is the preprocessed sample data set, and the feature extraction module is used to extract feature vectors from the sample data set.
[0059] In the present invention, the Transform-based feature extraction module includes a multi-head self-attention layer and a feedforward neural network layer, wherein the operation process of the self-attention layer includes: embedding the data in the preprocessed sample data set into the multi-head self-attention layer in the feature extraction module, and generating a query vector, a key vector and a value vector respectively; determining the attention score according to the query vector and the key vector using the Softmax function; determining the attention feature map according to the attention score and the value vector; and combining the attention feature maps to obtain a multi-head self-attention result.
[0060] In the embodiment of the present invention, the Transform-based feature extraction module includes two multi-head self-attention layers and two feedforward neural network layers, wherein the operation process of the multi-head self-attention layer is:
[0061] The production data in the pump inspection cycle in the normalized sample data set is embedded using formula (2) to generate the query vector Q, key vector K and value vector V respectively. The formulas are as follows:
[0062] (Q,K,V)=X i (W q ,W k ,W v ) (2);
[0063] Where, X i is the normalized production data, W q ,W k and W v is the weight matrix.
[0064] The method for determining the attention score by using the Softmax function according to the query vector and the key vector includes: multiplying the query vector Q and the key vector K and performing the Softmax function to obtain the attention score Attention(Q, K, V):
[0065]
[0066] The attention feature map Head is obtained by weighted summing the attention score and the value vector V i :
[0067]
[0068] Combine multiple attention feature maps to get the multi-head self-attention result MultiHead(Q,K,V):
[0069] MultiHead(Q,K,V)=Concat(Head i )W O (5);
[0070] In formulas (3) to (5), d k is the dimension of the key vector K, and W O are different weight matrices.
[0071] Among them, the feedforward neural network layer contains two layers of fully connected networks and uses the Relu activation function.
[0072] The preprocessed sample dataset is input into the Transform-based feature extraction module, where the input data is first converted into a high-dimensional vector representation for each word through an embedding layer. Then, position information is introduced into the vector representation by adding position encoding. Position encoding can be a fixed or learnable parameter. Next, the final feature vector is obtained by stacking a multi-head self-attention layer and a feedforward neural network layer.
[0073] Step S04: establishing a feature mapping module based on ResNet, wherein the input of the feature mapping module is the feature vector, and utilizing the feature mapping module to generate a pump inspection cycle prediction model for a pumping well based on Transform-ResNet.
[0074] In the present invention, the ResNet-based feature mapping module includes a residual module, a fully connected layer and a loss function, and the residual module includes a convolutional layer and an activation function;
[0075] Wherein, the loss function is:
[0076]
[0077] Where P j is the predicted pump inspection cycle value output by the model, Y j It is the actual pump inspection cycle value.
[0078] In the embodiment of the present invention, the input of the feature mapping module is the output of the feature extraction module in step S03. The feature mapping module based on ResNet includes 8 residual modules, 2 fully connected layers and a loss function, and the 8 residual modules include 16 convolutional layers and 8 activation functions.
[0079] ResNet is a deep residual network that is used to solve the problems of gradient vanishing and network degradation in deep neural networks. Its main function is to allow the training of deeper neural networks and improve the network's expressiveness and performance. One of the main problems of deep neural networks is gradient vanishing and gradient exploding. When the network becomes very deep, the gradient gradually decays or amplifies, resulting in ineffective backpropagation and training. This makes it difficult for earlier layers to learn effective feature representations, thus limiting the performance of the network.
[0080] ResNet solves this problem by introducing residual connections. Residual connections refer to adding the input directly to the output in the network so that the gradient can be directly transferred, thereby alleviating the problems of gradient vanishing and gradient exploding. Specifically, ResNet introduces residual blocks, adds the input to the output in each residual block, and then performs a nonlinear transformation through an activation function. In this way, the network can more easily learn the residual part, that is, the difference between the input and the expected output, without paying too much attention to learning the entire mapping.
[0081] The feature vector extracted by the feature extraction module is input into the feature mapping module based on ResNet, and a pump inspection cycle prediction model for pumping wells based on Transform-ResNet is generated through calculation.
[0082] Step S05: training and testing the pump inspection cycle prediction model according to the training set and the test set, and predicting the pump inspection cycle of the target pumping well by using the trained and tested pump inspection cycle prediction model.
[0083] In the embodiment of the present invention, the established pump inspection cycle prediction model for a pumping well based on the Transform-ResNet algorithm includes two parts: a feature extraction module based on Transform and a feature mapping module based on ResNet.
[0084] The training set data divided in step S02 is used to train the pump inspection cycle prediction model for the pumping unit well. The training set is a data sample used for model fitting. During the training process, the training error is gradient descended for learning. The test set data divided in step S02 is used to evaluate and test the model performance. If the performance does not meet the requirements, the training is re-performed by adjusting the parameters until the optimal model is screened out.
[0085] It can be understood that the above-mentioned various method embodiments mentioned in the present invention can be combined with each other to form a combined embodiment without violating the principle logic. Due to space limitations, the present invention will not go into details.
[0086] The execution subject of the method for predicting the pumping well inspection cycle may be a pumping well inspection cycle prediction device. For example, the method for predicting the pumping well inspection cycle may be executed by a terminal device or a server or other processing device, wherein the terminal device may be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the method for predicting the pumping well inspection cycle may be implemented by a processor calling a computer-readable instruction stored in a memory.
[0087] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.
[0088] The present invention also provides a pump inspection cycle prediction device for a pumping well, comprising: an acquisition unit, used to acquire existing pump inspection cycle production data of the pumping well, and establish a sample data set; a preprocessing unit, used to perform normalization preprocessing on all production data in the sample data set, and divide the normalized sample data set into a training set and a test set; a feature extraction unit, used to establish a feature extraction module based on Transform, the input of the feature extraction module is the preprocessed sample data set, and the feature extraction module is used to extract feature vectors in the sample data set; a model generation unit, used to establish a feature mapping module based on ResNet, the input of the feature mapping module is the feature vector, and the feature mapping module is used to generate a pump inspection cycle prediction model for the pumping well based on Transform-ResNet; a prediction unit, trained and tested the pump inspection cycle prediction model according to the training set and the test set, and predicted the pump inspection cycle of the target pumping well using the trained and tested pump inspection cycle prediction model for the pumping well.
[0089] In some embodiments, the functions or modules and units included in the device provided by the embodiment of the present invention can be used to execute the method described in the above method embodiment. Its specific implementation can refer to the description of the above method embodiment. For the sake of brevity, it will not be repeated here.
[0090] The present invention first obtains the production index data of the pump inspection cycle of the existing oil pumping wells to form a sample data set; then normalizes all the production data in the sample data set so that the preprocessed data set can be used for subsequent network training; then a Transform-based feature extraction module is designed to learn the features of the data, and Transform can extract the global features of the data, and a Resnet-based feature mapping module is used to map the data features to the pump inspection cycle, so that the Resnet shallow network can better learn the local features without losing the global features.
[0091] Compared with the traditional method for predicting the pump inspection cycle of a pumping well, the present invention has the following advantages and beneficial effects: 1. The present invention designs a Transform-Resnet algorithm model, which integrates the Transform algorithm with the Resnet algorithm, making up for the shortcoming of the Resnet algorithm in insufficient global feature extraction in shallow networks, and realizes end-to-end processing of pumping well production data, thereby improving the accuracy and adaptability of the model; 2. The present invention takes into account the noise of the field data, introduces a 10% error in the loss function design process, reduces the sensitivity of the model to noise, and improves the robustness of the system.
[0092] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.
Claims
1. A method for predicting the pump inspection cycle of a well pumping unit, characterized in that: include: Obtain existing production data of pump inspection cycles of oil wells and establish a sample data set; Performing normalization preprocessing on all production data in the sample data set, and dividing the normalized sample data set into a training set and a test set; Establishing a Transform-based feature extraction module, the input of the feature extraction module is the preprocessed sample data set, and using the feature extraction module to extract feature vectors in the sample data set; Establishing a feature mapping module based on ResNet, wherein the input of the feature mapping module is the feature vector, and using the feature mapping module to generate a pump inspection cycle prediction model for a pumping well based on Transform-ResNet; The pump inspection cycle prediction model is trained and tested according to the training set and the test set, and the pump inspection cycle of the target pumping well is predicted using the trained and tested pump inspection cycle prediction model for the pumping well.
2. The method for predicting the pump inspection cycle of a well pumping unit according to claim 1, characterized in that: The existing production data of the pump inspection cycle of the oil well pumping unit is: The pump inspection cycle production data of the pumping wells that are in the same block as the target block or have the same reservoir geological conditions as the target block and the same production conditions.
3. The method for predicting the pump inspection cycle of a well pumping unit according to claim 1, characterized in that: The production data at least includes: Daily liquid production, daily oil production, upstream current, downstream current, oil pressure, casing pressure, water content, stroke, number of strokes, pump depth and pump diameter.
4. The method for predicting the pump inspection cycle of a well pumping unit according to claim 1, characterized in that: The formula for the normalization process includes: In the formula, is the normalized production data, x is the original production data of the pumping well, X min is the minimum value in the production data, X max This is the maximum value in the production data.
5. The method for predicting the pump inspection cycle of a well pumping unit according to claim 1, characterized in that: Before the normalization process, the method further includes: If there are gaps in the obtained production data of the pump inspection cycle of the existing pumping well, the missing data are filled by taking the average value of the predetermined number of data before and after the missing data.
6. The method for predicting the pump inspection cycle of a well pumping unit according to claim 1, characterized in that: The Transform-based feature extraction module includes a multi-head self-attention layer and a feedforward neural network layer, wherein the operation process of the self-attention layer includes: Embed the data in the preprocessed sample data set into the multi-head self-attention layer in the feature extraction module to generate a query vector, a key vector and a value vector respectively; Determine an attention score using a Softmax function according to the query vector and the key vector; Determining an attention feature map according to the attention score and the value vector; The attention feature maps are combined to obtain multi-head self-attention results.
7. The method for predicting the pump inspection cycle of a well pumping unit according to claim 1, characterized in that: The ResNet-based feature mapping module includes a residual module, a fully connected layer and a loss function, and the residual module includes a convolutional layer and an activation function; Wherein, the loss function is: Where P j is the predicted pump inspection cycle value output by the model, Y j It is the actual pump inspection cycle value.
8. A pump inspection cycle prediction device for a well pumping unit, characterized in that: include: An acquisition unit, used to acquire existing production data of the pump inspection cycle of the pumping well and establish a sample data set; A preprocessing unit, used for performing normalization preprocessing on all production data in the sample data set, and dividing the normalized sample data set into a training set and a test set; A feature extraction unit, used to establish a Transform-based feature extraction module, the input of the feature extraction module is the preprocessed sample data set, and the feature extraction module is used to extract feature vectors in the sample data set; A model generation unit is used to establish a feature mapping module based on ResNet, the input of the feature mapping module is the feature vector, and the feature mapping module is used to generate a pump inspection cycle prediction model for a pumping well based on Transform-ResNet; The prediction unit trains and tests the pump inspection cycle prediction model according to the training set and the test set, and predicts the pump inspection cycle of the target pumping well using the trained and tested pump inspection cycle prediction model.