A method for predicting the displacement of an electric pump well based on power method analysis and BP network modeling

Abstract

The present application relates to the technical field of intelligent oilfield electric submersible pump, and particularly relates to an electric pump well displacement prediction method based on power method analysis and BP network modeling, comprising the following steps: S1, using the power method, using the system energy conservation theorem, analyzing the system parameters related to the electric submersible pump well yield; S2, recording the data parameters of the electric pump well analyzed in S1, and collecting the corresponding volume displacement of the electric submersible pump well; S3, preprocessing the data parameters and volume displacement data of the electric pump well; S4, using the neuron net input, designing the corresponding multivariate function model; S5, establishing the BP neural network based on the multivariate function model, and obtaining the BP network model through iterative training; S6, using the trained BP network model to predict the electric submersible pump well yield. According to the electric submersible pump well displacement prediction model, the wellhead displacement can be calculated in real time, and the error rate compared with the artificial metering result is less than 10%, which can meet the demand of field application.

Description

Technical Field

[0001] This invention relates to the field of intelligent oilfield electric submersible pump technology, and in particular to a method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling. Background Technology

[0002] ESP wells, also known as submersible pump wells, are widely used in deviated wells, directional wells, and high-yield wells due to their compact structure and high efficiency. The production volume of ESP wells is a key indicator for analyzing their operational status. Due to the limitations of offshore oil platforms, the production volume of ESP wells is manually measured periodically using separator glass tube metering equipment. When weather conditions are severe, metering personnel cannot go to sea, and ESP well production often remains unmeasured for half a month or even longer. ESP wells are the primary production method for offshore oil platforms, with 95% of the wells in the plant using ESP wells. Failure to measure wellhead production in a timely manner will severely hinder the production management of offshore platforms. Researching ESP well metering methods to reduce offshore oilfield production costs has become an important research direction in offshore oil development.

[0003] In recent years, Shengli Oilfield has continuously carried out the "Four Modernizations" construction, accumulating a large amount of data assets for its various management areas. Automatic acquisition and low-latency transmission of operating parameters for submersible electric pump wells in offshore oilfields have been achieved; parameters such as wellhead oil pressure, casing pressure, wellhead back pressure, wellhead temperature, three-phase current, and three-phase voltage are automatically acquired; some wells achieve real-time monitoring of bottom hole flowing pressure data through capillary steel tube pressure; and the interval between data acquisition for operating parameters is approximately 5 minutes. Historical data on operating parameters of electric pump wells dating back to 2000 exceeds 3.5 billion records, representing abundant and rapidly updated data resources that urgently need to be explored and utilized. With the rapid development of data mining, artificial intelligence, and other technologies, various neural network models are emerging in large numbers. How to model existing data, select suitable deep neural networks for training, and generate an automatic prediction model for submersible electric pump production is a pressing problem that needs to be solved. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling.

[0005] The technical solution is as follows:

[0006] A method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling includes the following steps:

[0007] S1. Using the power method and the law of conservation of system energy, we analyze the system parameters related to the production of submersible electric pump wells.

[0008] S2 records the data parameters of the electric pump well analyzed in S1 and collects the corresponding volumetric displacement of the submersible electric pump well.

[0009] S3, preprocesses the data parameters and volumetric displacement data of the electric pump well;

[0010] S4, using Net in Based on the net input to neurons, design the corresponding multivariate function model;

[0011] S5, a BP neural network is established based on a multivariate function model, and the BP network model is obtained through iterative training;

[0012] S6 uses a trained BP network model to predict the production of submersible electric pump wells.

[0013] Furthermore, the feature is that, for step S1, the submersible electric pump well is considered as a closed system, which then conforms to the law of conservation of energy, thus the following identity exists:

[0014] Input electrical energy + formation flow energy = lift liquid potential energy + frictional loss energy (1)

[0015] Lifting liquid potential energy = Input electrical energy + Formation flow energy - Frictional loss energy (2)

[0016] Therefore, the wellhead displacement Q can be expressed as a function of the following formula:

[0017] Q=f(U,I,ρ 产出液 H 泵 ,P 油 ,T 井口 (3)

[0018] In equation (3), U represents the three-phase voltage; I represents the three-phase current; ρ 产出液 The density of the fluid produced by the oil well is expressed by the following formula:

[0019] ρ 产出液 =ρ w ×f w +ρ0×(1-f w (4)

[0020] In equation (4), ρ w ρ0 represents the density of water produced at the bottom of the well, which can be obtained through laboratory analysis; ρ0 represents the density of crude oil at the surface, which can be obtained through analytical testing data; f w Indicates the water content of the product liquid; H 泵 Indicates the pump hanging depth; P 油 Indicates wellhead oil pressure; T 井口 This indicates the wellhead temperature.

[0021] Furthermore, the feature is that, for step S2, U, I, P 油 T 井口Intelligent automated data collection has been achieved, gathering data once per minute and storing it in a database; ρ 产出液 It can be obtained from the analysis of crude oil on the ground; H 泵 The displacement Q of submersible electric pump wells can be approximated by manually measured data and can be obtained from previous operational data.

[0022] Furthermore, the feature is that, for step S3, since the operating environment parameters of the submersible electric submersible pump are affected by field instruments and network transmission, they are generally incomplete, inconsistent, and dirty data, which cannot be used directly and require the following processing:

[0023] a. Abnormal data removal

[0024] There is a reasonable range of operating parameters for submersible electric submersible pumps. Data is filtered according to the range, and parameters that do not conform to the range are deleted.

[0025] b. Data interpolation completion

[0026] Linear interpolation was used to fill in missing data.

[0027] Furthermore, the feature is that, for step S4, the input is the key influencing variables of the model obtained in the above steps; there are multiple ways to combine the signal with the input to the neuron, through Net in A multivariate function model is established based on the net input of neurons:

[0028]

[0029] Furthermore, the feature is that step S5 includes the following empirical formula:

[0030]

[0031] In equation (6), n is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is a constant between [2, 12].

[0032] The initial weights are selected between (-1, 1), and the learning rate is between (-0.01, 0.8).

[0033] Before inputting the data into the BP network, the data is normalized using a linear function transformation method, as shown in the following expression:

[0034]

[0035] In equation (7), x and y are the values ​​before and after normalization, respectively, and Max and Min are the maximum and minimum values ​​of the parameter, respectively.

[0036] Furthermore, the feature is that an sigmoid transfer function is used to express the nonlinearity of the network, and its expression is as follows:

[0037]

[0038] The error function E is minimized by continuously adjusting the network weights and thresholds using the backpropagation error function, and its expression is as follows:

[0039]

[0040] In equation (9), t i Indicates the expected output, O i This represents the computational output of the network.

[0041] Once the network training is complete, inputting the various parameters into the trained BP network model will yield the predicted discharge rate of the electric pump well.

[0042] The beneficial effects of this invention are as follows:

[0043] The submersible electric pump well displacement prediction model can realize real-time wellhead displacement calculation, and the error rate compared with manual measurement results is less than 10%, which can meet the needs of field applications. Attached Figure Description

[0044] Figure 1 This is a flowchart of the electric pump well discharge prediction method based on power method analysis and BP network modeling in this invention.

[0045] Figure 2 This is the BP network structure of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0047] Example 1:

[0048] A method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling, the specific steps of which are as follows:

[0049] S1. Using the power method and the law of conservation of system energy, we analyze the system parameters related to the production of submersible electric pump wells.

[0050] S2 records the data parameters of the electric pump well analyzed in S1 and collects the corresponding volumetric displacement of the submersible electric pump well.

[0051] S3, preprocesses the data parameters and volumetric displacement data of the electric pump well;

[0052] S4, using Netin Based on the net input to neurons, design the corresponding multivariate function model;

[0053] S5, a BP neural network is established based on a multivariate function model, and the BP network model is obtained through iterative training;

[0054] S6. Predict the production of submersible electric pump wells using a trained BP network model;

[0055] The technical features and improvements of this invention are as follows:

[0056] For step S1, if we consider the submersible pump well as a closed system, then this system conforms to the law of conservation of energy, and thus the following identity exists:

[0057] Input electrical energy + formation flow energy = lift liquid potential energy + frictional loss energy (1)

[0058] Lifting liquid potential energy = Input electrical energy + Formation flow energy - Frictional loss energy (2)

[0059] The input electrical energy can be calculated using the three-phase voltage and current of the motor input; frictional loss energy includes the sum of all losses from the motor and friction; the two operating parameters, oil pressure and temperature, have a functional relationship with "formation flow energy - frictional loss energy". Therefore, it can be deduced that the displacement Q of a submersible electric pump well is related to the wellhead voltage, wellhead current, produced fluid density, bottom hole flowing pressure, oil pressure, and temperature. Thus, the function of the wellhead displacement Q can be expressed as the following formula:

[0060] Q=f(U,I,ρ 产出液 H 泵 ,P 油 ,T 井口 (3)

[0061] In equation (3), U represents the three-phase voltage; I represents the three-phase current; ρ 产出液 The density of the fluid produced by the oil well is expressed by the following formula:

[0062] ρ 产出液 =ρ w ×f w +ρ0×(1-f w (4)

[0063] In equation (4), ρ w ρ0 represents the density of water produced at the bottom of the well, which can be obtained through laboratory analysis; ρ0 represents the density of crude oil at the surface, which can be obtained through analytical testing data; f w Indicates the water content of the product liquid; H 泵 Indicates the pump hanging depth; P 油 Indicates wellhead oil pressure; T 井口 This indicates the wellhead temperature.

[0064] For step S2, U, I, P 油 T 井口 Intelligent automated data collection has been achieved, gathering data once per minute and storing it in a database; ρ 产出液 It can be obtained from the analysis of crude oil on the ground; H 泵 It can be obtained from past operational data. The displacement Q of a submersible electric pump well can be approximated as manually measured data.

[0065] For step S3, the operating environment parameters of the submersible electric submersible pump are largely incomplete, inconsistent, and corrupted due to the influence of field instruments and network transmission, and cannot be used directly. The following processing is required:

[0066] a. Abnormal data removal

[0067] Submersible electric submersible pumps (ESPs) have reasonable operating parameter ranges. Data can be filtered based on these ranges, and parameters that do not fall within the range can be deleted. For example, three-phase voltage less than 796V, three-phase current less than 0A, temperature greater than 100℃, and oil pressure lower than back pressure can all be filtered out using data cleaning programs.

[0068] b. Data interpolation completion

[0069] Ideally, the operating parameters of the electric pump well are collected every five minutes. However, two factors can lead to missing data: first, data gaps caused by failures in front-end instrument acquisition or transmission; and second, data gaps during the abnormal data cleaning process. To facilitate subsequent steps, missing data needs to be filled in. This invention employs linear interpolation.

[0070] For step S4, the input consists of the key influencing variables of the model obtained in the previous steps; there are multiple ways to combine signals with inputs to neurons, and this invention selects Net. in A multivariate function model is established based on the net input of neurons.

[0071]

[0072] For step S5, this invention designs a three-layer multi-input single-output BP network prediction model with one hidden layer. In the BP network design process, the number of neurons in the hidden layer plays a crucial role in the model quality. Too many hidden layer neurons will increase the computational cost of the BP network and lead to overfitting; too few neurons will result in poor network fitting and the model will fail to meet prediction accuracy requirements. Currently, there is no clear definition of how to select the number of hidden layer neurons; only some empirical formulas exist. This invention referenced the following empirical formulas in the initial stage:

[0073]

[0074] In equation (6), n is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is a constant between [2, 12].

[0075] Initial weights are typically chosen between (-1, 1). The learning rate is generally between (-0.01, 0.8). A learning rate that is too large may lead to instability during training, while a learning rate that is too small may increase the model training time. To reduce model training time, this invention employs a variable adaptive learning rate, allowing the network to have different learning rates at different stages. The desired error needs to be determined through multiple comparative training iterations to find an appropriate value, which is determined relative to the number of hidden layer nodes. This invention simultaneously trains two networks with different desired error values, and finally determines the desired error based on performance.

[0076] Before inputting data into the BP network, the data needs to be normalized. This invention uses a linear function transformation method, the expression of which is as follows:

[0077]

[0078] In equation (7), x and y are the values ​​before and after normalization, respectively, and Max and Min are the maximum and minimum values ​​of the parameter, respectively.

[0079] This invention uses a sigmoid transfer function to express the nonlinearity of the network, and its expression is as follows:

[0080]

[0081] The error function E is minimized by continuously adjusting the network weights and thresholds using the backpropagation error function, and its expression is as follows:

[0082]

[0083] In equation (9), t i Indicates the expected output, O i This represents the computational output of the network.

[0084] Set the number of network iterations (epochs) to 5000 and the expected error to 10. -8 The network is trained, and training is completed after repeated learning until the desired error is reached.

[0085] Once the network training is complete, inputting the various parameters into the trained BP network model will yield the predicted discharge rate of the electric pump well.

[0086] Example 2:

[0087] A method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling, such as Figure 1 The diagram shown is a flowchart of the electric pump well displacement prediction method based on power method analysis and BP network modeling of the present invention. The method includes:

[0088] S1, Parameter Correlation Analysis: By using the power method, the submersible electric pump well is regarded as a closed system. The energy inside the system is conserved. By analyzing the law of conservation of energy, the pump well's discharge capacity is mainly related to the three-phase voltage, three-phase current, well produced fluid density, pump hanging depth, and wellhead oil pressure parameters, which facilitates the analysis in the following steps.

[0089] S2, the sample set for collecting data includes three-phase voltage, three-phase current, wellhead oil pressure, and wellhead temperature, all collected automatically once per minute. The density of the produced fluid is obtained through analysis of the crude oil on the surface. The pump hanging depth is obtained by retrieving data from previous downhole operations. The wellhead discharge rate can be approximated as manually measured data.

[0090] S3: Due to the influence of field instruments and network transmission on the operating environment parameters of the electric submersible pump, the data is generally incomplete, inconsistent, and dirty, making it unusable directly. It is necessary to remove outliers and interpolate to complete the raw data. In addition, the processed data needs to be normalized to standardize its format.

[0091] S4, the input is the key influencing variables of the model obtained in the above steps; there are multiple ways to combine signals with input to neurons, and this invention selects Net. in A multivariate function model is established based on the net input of neurons.

[0092] S5, the initial weights are selected as random numbers between (-1, 1). The learning rate is between (-0.01, 0.8). A learning rate that is too large may lead to instability during training, while a learning rate that is too small may increase the model training time. To reduce model training time, this invention employs a variable adaptive learning rate, allowing the network to have different learning rates at different stages. A sigmoid transfer function is used to represent the nonlinearity of the network; the backpropagation error function is used to continuously adjust the network weights and thresholds to minimize the error function E. This is how the BP network structure is established, as follows: Figure 2 As shown.

[0093] S6 can make predictions through the established BP network prediction model, realizing the prediction of the discharge of submersible electric pump wells under the current state.

[0094] In summary, the electric pump well displacement prediction method based on power method analysis and BP network modeling of the present invention provides a new approach to electric pump well production measurement, and solves the inherent problems of manual production measurement such as long single-well measurement cycle and low production measurement efficiency.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims.

Claims

1. A method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling, characterized in that, The process includes the following steps: S1, using the power method and the law of conservation of system energy, analyze the system parameters related to the production of submersible electric pump wells; S2, record the data parameters of the electric pump wells analyzed in S1, and collect the corresponding volumetric displacement of the submersible electric pump wells; S3, preprocess the data parameters and volumetric displacement data of the electric pump wells; S4, use Net... in S5: Based on the net input of neurons, design the corresponding multivariate function model; S6: Based on the multivariate function model, establish a BP neural network and obtain the BP network model through iterative training; S7: Use the trained BP network model to predict the production of submersible electric pump wells. For step S1, if we consider the submersible pump well as a closed system, then this system conforms to the law of conservation of energy, and thus the following identity exists: Input electrical energy + formation flow energy = lift liquid potential energy + frictional loss energy (1) Lifting potential energy of liquid = Input electrical energy + Formation flow energy - Frictional loss energy (2) Therefore, the wellhead displacement Q can be expressed as a function of the following formula: Q=f(U,I,ρ 产出液 ,H 泵 ,P 油 ,T 井口 ) (3) In equation (3), U represents the three-phase voltage, I represents the three-phase current, and H represents the three-phase current. 泵 Indicates pump hanging depth, P 油 Indicates wellhead oil pressure, T 井口 ρ represents the wellhead temperature. 产出液 The density of the fluid produced by the oil well is expressed by the following formula: In equation (4), ρ w ρ0 represents the density of water produced at the bottom of the well, which can be obtained through laboratory analysis; ρ0 represents the density of crude oil at the surface, which can be obtained through analytical testing data; f w This indicates the water content of the produced liquid.

2. The method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling as described in claim 1, characterized in that, For step S2, U, I, P 油 T 井口 Intelligent automated data collection has been achieved, gathering data once per minute and storing it in a database; ρ 产出液 It can be obtained from the analysis of crude oil on the ground; H 泵 The displacement Q of submersible electric pump wells can be approximated by manually measured data and can be obtained from previous operational data.

3. The method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling according to claim 2, characterized in that, For step S3, since the operating environment parameters of the submersible electric submersible pump are affected by field instruments and network transmission, they are generally incomplete, inconsistent, and dirty data, which cannot be used directly. The following processing is required: a. Abnormal data removal There is a reasonable range of operating parameters for submersible electric submersible pumps. Data is filtered according to the range, and parameters that do not conform to the range are deleted. b. Data interpolation completion Linear interpolation was used to fill in missing data.

4. The method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling according to claim 3, characterized in that, For step S4, the input consists of the key influencing variables of the model obtained in the previous steps; there are multiple ways to combine signals with inputs to neurons, such as through a Net. in A multivariate function model is established based on the net input of neurons: (5); In the formula, w i Let x represent the weight coefficient of the i-th input. i This represents the i-th input variable.

5. The method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling according to claim 4, characterized in that, For step S5, the following empirical formula is included: 6); In equation (6), I is the number of hidden layer neurons, n is the number of input layer neurons, m is the number of output layer neurons, and a is a constant between [2, 12]; the initial weights are selected between (-1, 1), and the learning rate is between (-0.01, 0.8). Before inputting the data into the BP network, the data is normalized using a linear function transformation method, as shown in the following expression: (7) In equation (7), x and y are the values ​​before and after normalization, respectively, and Max and Min are the maximum and minimum values ​​of the parameter, respectively.

6. The method for predicting the discharge capacity of electric pump wells based on power method analysis and BP network modeling according to claim 5, characterized in that, The sigmoid transfer function is used to express the nonlinearity of the network, and its expression is as follows: (8) In the formula, x is the net input value of the hidden layer neuron. The error function E is minimized by continuously adjusting the network weights and thresholds through the backpropagation error function, and its expression is as follows: (9); In equation (9), t i Indicates the expected output, O i This represents the computational output of the network, where i is the training sample number. After the network training is completed, inputting the various parameters into the trained BP network model will yield the predicted discharge rate of the electric pump well.

Images