High-precision testing device and interpretation method for water content of oil well produced liquid based on intelligent modeling

Abstract

The invention provides an intelligent modeling-based high-precision testing device and interpretation method for the water content of oil well produced fluid, and the method comprises the steps: 1, carrying out the data preprocessing, and screening main factors affecting the water content through a PCA method; step 2, performing maximum-minimum normalization processing to form a water content influence factor feature vector; 3, dividing the feature vectors into a training set and a test set; 4, creating a sliding window, and constructing and training an LSTM water content prediction model; 5, training parameters are optimized, and a training result is evaluated; 6, testing and verifying the LSTM water content prediction model; and step 7, carrying out future short-term prediction on the water content of the LSTM. According to the high-precision testing and interpretation method for the water content of the oil well produced liquid based on intelligent modeling, the water content is accurately monitored, meanwhile, the flow is accurately interpreted, and the problem of produced liquid metering errors caused by the fact that a traditional water content testing technology and a flow testing technology are not monitored at the same time is solved.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field development technology, and in particular to a high-precision testing device and interpretation method for water content in oil well production fluids based on intelligent modeling. Background Technology

[0002] Currently, with the deep development of oilfields, most domestic oilfields have entered the mature stage. The produced fluids are mostly two-phase mixtures of crude oil and water, and the water cut is constantly increasing. Some oilfields have even entered the ultra-high water cut stage (water cut greater than 90%). Oil well production fluid measurement is a crucial indicator in oilfield development, serving as an important basis for evaluating the development status and designing and adjusting development plans. This includes measuring water cut, gas cut, and total fluid volume. Among these, determining the water cut of crude oil allows for understanding and monitoring the production status of each well and each producing layer, enabling real-time monitoring of crude oil properties. This allows for targeted fracturing or water injection of underground reservoirs, guiding subsequent production and refining processes, and improving well recovery rates and oilfield production efficiency. Therefore, measuring the water cut of crude oil is of great practical significance for crude oil extraction. Currently, water cut testing in domestic oil wells mainly relies on manual sampling and analysis, with samples typically taken every 2-3 days. Production fluid measurement is primarily based on dynamometer card analysis, usually performed every half hour. Production fluid cannot be tested simultaneously with water cut, which severely affects the accuracy of production fluid flow rate measurement.

[0003] Currently, commonly used water cut measurement methods both domestically and internationally are mainly divided into manual measurement methods and online measurement methods. Typical manual measurement methods include distillation, electrostatic desorption, and Karl Fischer methods. Manual measurement methods offer relatively high accuracy, but the results are random and not timely. Sampling and measurement processes are time-consuming and labor-intensive, failing to meet the needs of oilfield field exploration and automated production management. In recent years, online measurement has become a development trend. Online measurement methods mainly include X-ray methods, capacitance methods, conductivity methods, microwave methods, radio frequency methods, and coaxial phase methods. However, online measurement technology is susceptible to environmental influences, has poor adjustability, a small measurement range, and low accuracy.

[0004] Existing online water cut testing technology for oil wells using dual-rectifier systems with in-pipe phase separation employs modular configurations to achieve online monitoring of water cut. The advantages of this method are that the testing system can continuously and automatically monitor changes in produced fluid water cut 24 hours a day; it is unaffected by water salinity and has strong adaptability to different oil products; and it has wireless remote transmission capabilities, enabling real-time monitoring of oil well production dynamics. However, the data generated by this method only uses an average value interpretation to determine the current water cut, resulting in low data processing accuracy. Furthermore, it cannot predict water cut changes over a period of time, making it difficult to promptly address and control situations that occur in the short term. Therefore, there is an urgent need for a simple and practical method that can improve the accuracy of current water cut data processing and the accuracy of water cut prediction over a future period.

[0005] In recent years, with the continuous development of computer and internet technologies, machine learning algorithms in the big data environment have been gradually applied to the oil and gas industry. Machine learning algorithms such as Support Vector Machine (SVM), Artificial Neural Network (ANN), Convolutional Neural Network (CNN), and Long Short-Term Memory Neural Network (LSTM) are commonly used in wellbore multiphase flow characterization studies. Among them, the LSTM algorithm, due to its long-term and short-term memory properties, has a greater advantage in predicting long-term or short-term time series data in the field of oil and gas field development.

[0006] Patent CN202110715729.7 mainly provides a method and system for rapid prediction of single-well production using a constructed LSTM model; in 2020, Wang Hongliang et al. published a method for predicting oilfield production during the ultra-high water cut period based on recurrent neural networks in the field of oil exploration and development. Although these patents and articles only introduce the method of using LSTM models for production prediction, they also demonstrate the feasibility of LSTM models for long-term or short-term prediction in the oil and gas field, proving that LSTM models can be used for oilfield water cut interpretation.

[0007] Patent CN201910613947.2 mainly provides a system and method for predicting the water cut at the wellhead of low-yield gas and oil wells based on a deep long short-term memory network. However, the data used in this patent is the water cut fluctuation information of the wellhead mixture obtained by a high-frequency dual-ring capacitive sensor, which is unrelated to the online water cut testing technology for dual-rectifier oil wells based on in-pipe phase separation used in this invention.

[0008] In summary, current oil well water cut measurement technologies suffer from accuracy and real-time performance issues. Utilizing dual-rectifier oil well water cut testing technology based on in-pipe phase separation allows for online water cut measurement. Machine learning algorithms offer significant advantages in data processing. Improving the accuracy of current water cut data processing and future water cut prediction using machine learning algorithms has been a key technical focus in this field, holding practical significance for oilfield development. Therefore, we have invented a novel high-precision oil well water cut testing device and interpretation method based on intelligent modeling. Summary of the Invention

[0009] The purpose of this invention is to provide a high-precision testing device and interpretation method for water content in oil well production based on intelligent modeling, which overcomes the shortcomings of low accuracy and narrow applicability of traditional indoor experimental calibration of water content models and improves the accuracy of water content data processing.

[0010] The objective of this invention can be achieved through the following technical measures: a high-precision testing device for water cut in oil well production fluid based on intelligent modeling. This device includes a connecting body, a vortex shaping and differential pressure generation section, a test data monitoring section, and a data acquisition and intelligent analysis section. The vortex shaping and differential pressure generation section is connected to the connecting body, which ensures stable and steady production fluid flow. The vortex shaping and differential pressure generation section shapes the production fluid to generate radial and axial differential pressures. The test data monitoring section is connected to the vortex shaping and differential pressure generation section, which collects the radial differential pressure and friction resistance of the production fluid and transmits them to the data acquisition and intelligent analysis section. Based on the received data, the data acquisition and intelligent analysis section uses an LSTM (Long Short-Term Memory) neural network algorithm to perform flow rate and water cut test analysis.

[0011] The objective of this invention can also be achieved through the following technical measures:

[0012] The main body of the connection includes an inlet section, a flow stabilizing pipe section, an upper connecting flange, an upper axial differential pressure chamber, a lower axial differential pressure chamber, an upper radial differential pressure chamber, a lower radial differential pressure chamber, a main body sealing chamber, and a lower connecting flange. The inlet section is connected to the flow stabilizing pipe section to ensure a stable and steady flow rate of the incoming product liquid. The upper connecting flange is connected to the flow stabilizing pipe section. The upper connecting flange is sequentially connected to the upper axial differential pressure chamber, the upper radial differential pressure chamber, the main body sealing chamber, the lower radial differential pressure chamber, and the lower axial differential pressure chamber to form the main body of the main body. Then, it is connected to the lower connecting flange and leads to the outlet.

[0013] The swirl shaping and differential pressure generation section includes an upper swirl shaper, a middle swirl radial differential pressure generator, a lower swirl shaper, an axial differential pressure tapping bend, and a radial differential pressure tapping straight pipe. The upper swirl shaper, the middle swirl radial differential pressure generator, and the lower swirl shaper are connected sequentially, forming a stable annular flow with a water ring on the outside and an oil core in the center within the dual rectifier. The measured axial and radial differential pressures are transmitted to the test data monitoring section through the axial differential pressure tapping bend and the radial differential pressure tapping straight pipe.

[0014] The test data monitoring section includes an inlet pressure sensor, an outlet temperature sensor, a radial differential pressure sensor, and an axial differential pressure sensor. The inlet pressure sensor collects the pressure data of the product liquid, the outlet temperature sensor collects the temperature data of the product liquid, the radial differential pressure sensor collects the radial differential pressure generated by the intermediate swirling radial differential pressure generator, and the axial differential pressure sensor collects the friction resistance of the upper swirling shaper, the intermediate swirling radial differential pressure generator, and the lower swirling shaper.

[0015] The data acquisition and intelligent analysis section receives data transmitted from the test data monitoring section and calculates the relationship between axial pressure difference and flow rate as follows:

[0016]

[0017] Where, ΔP z For axial pressure difference; α z ρ is the flow coefficient corresponding to the axial pressure difference; Q is the flow rate; A is the cross-sectional area of ​​the pipe; ρ is the density of the fluid.

[0018] The objective of this invention can also be achieved through the following technical measures: a high-precision testing and interpretation method for water cut in oil well production based on intelligent modeling. This method employs a high-precision testing device for water cut in oil well production based on intelligent modeling, comprising:

[0019] Step 1: Perform data preprocessing and use PCA method to screen the main factors affecting water content;

[0020] Step 2: Perform max-min normalization to form the feature vector of water content influencing factors;

[0021] Step 3: Divide the feature vectors into a training set and a test set;

[0022] Step 4: Create a sliding window, build an LSTM water content prediction model, and train it;

[0023] Step 5: Optimize training parameters and evaluate training results;

[0024] Step 6: Test and validate the LSTM water content prediction model;

[0025] Step 7: Perform LSTM short-term prediction of future water content.

[0026] The objective of this invention can also be achieved through the following technical measures:

[0027] In step 1, a time series dataset is formed based on the data collected by the high-precision water cut testing device for oil well production fluid based on intelligent modeling, and the main influencing factors of water cut are analyzed. Factors related to water cut include radial pressure difference, axial pressure difference, pressure difference ratio (axial pressure difference / radial pressure difference ZR), production fluid temperature, water cut factor variation, flow coefficient, measured flow rate, and structural parameters of the cyclone device, including the pipe diameter of the measuring device, the hydrocyclone angle, the hydrocyclone rotation angle, and the pressure tapping method.

[0028] In step 1, if there are many factors affecting water content, more than 10, then principal component analysis (PCA) is used to screen out the 5-6 factors most relevant to water content.

[0029] In step 2, the water content influencing factors after PCA analysis in step 1 are subjected to maximum-minimum normalization to establish a standard machine learning dataset. The maximum-minimum normalization transformation formula is:

[0030]

[0031] Among them, X norm X represents the result after data normalization; X represents the original data; X max X represents the maximum value in the data; min This represents the minimum value in the data.

[0032] Construct a feature vector sample set using the normalized data; assume X t Let Y be the feature vector of water-bearing influencing factors at time t, and let Y be the predicted water-bearing value at time t. t Each feature vector contains n features, n≤10, numbered F1-F1. n .

[0033] In step 3, the sample set processed in step 2 is processed according to a selected time t. o Divide into, t o Time and t o The sample data before time step t is used as the training set. o The sample data after the specified time is used as the test set.

[0034] In step 4, create a sliding window, set the window time step to Δt, and input the feature vector X for the first Δt time interval. t To predict the water content at time Δt+1, the time step is shifted by Δt each time to predict the water content at the next time Δt+1, until the entire training set is iterated.

[0035] In step 4, the sample set used by the LSTM model consists of the input time series and the output time series. The input time series is the feature vector X of water content influencing factors. t X t It contains n features F1-F n The output time series is the predicted water content Y at time t. t The training set after step 3 is input into the constructed LSTM model. The training set is processed through the input gate, forget gate and output gate to obtain the output of the first LSTM layer. The output vector of the first LSTM layer is used as the input vector of the second LSTM layer, and so on, iterating continuously. The output of each LSTM layer is the input of the next layer.

[0036] In step 5, to evaluate the accuracy of the LSTM model in water content prediction, the optimal coefficient of determination R is selected. 2The three evaluation indicators are: Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE); the specific relationship is as follows:

[0037]

[0038]

[0039]

[0040] Among them, y i This is the actual value. The average of the actual values. These are the predicted values ​​from the LSTM model;

[0041] The quality of LSTM model training is judged by evaluation metrics. When the metric value is less than the given error value, the model training is good and meets the requirements for liquid water content prediction, and an LSTM water content prediction model can be formed. Otherwise, the model performance is poor and needs to be retrained until it meets the requirements.

[0042] In step 6, the test set divided in step 3 is input into the LSTM water content prediction model trained and optimized in steps 4 and 5 for prediction; after the prediction is completed, the predicted water content value is output, and the output predicted value is inversely normalized; the predicted water content value is compared with the actual sampled water content value to analyze whether the model prediction effect meets the field requirements.

[0043] In step 7, prepare a set of data collected through high-precision testing of water content in oil well production based on intelligent modeling. Process the data according to steps 1 and 2 to form a feature vector of water content influencing factors. Input the feature vector into the LSTM water content prediction model trained, optimized and tested in steps 4, 5 and 6, and output the predicted water content value in the short term.

[0044] The high-precision testing device and interpretation method for water cut in oil well production based on intelligent modeling in this invention is a high-precision testing and interpretation method for water cut in oil well production based on principal component analysis (PCA) and long short-term memory neural network (LSTM) models. The technical advantages of this invention are:

[0045] (1) This invention utilizes the dual-rectifier oil well water cut testing technology based on in-pipe phase separation to achieve real-time online monitoring of oil well water cut. The monitoring system has excellent adaptability to rotating flow and eddy current conditions, and the measured values ​​can accurately and sensitively reflect the water cut changes of the measured oil well's produced fluid. This is of great significance for real-time monitoring of oil well production dynamics and guiding oilfield optimization management.

[0046] (2) This invention solves the data processing problem collected from the water cut testing technology of dual-rectifier oil wells. It uses PCA technology to analyze multiple factors affecting water cut and screens the factors most related to water cut. This can improve the accuracy of data selection, reduce the data dimension, and speed up the calculation. At the same time, it provides a more accurate and targeted feature for the establishment of water cut prediction model.

[0047] (3) The constructed LSTM neural network model can predict the water content of oil well production with high accuracy, which makes it easier for the field to understand the water content changes that will occur in the near future, and facilitates the targeted implementation and control of field measures, thus guiding field production work.

[0048] (4) This invention realizes the optimization of structural parameters of the dual-rectifier oil well water cut testing system and the prediction of oil well production water cut using the LSTM model. The process is rigorous and practically applicable. Compared with traditional water cut measurement and prediction methods, this invention has the advantages of being unaffected by environmental interference, high measurement accuracy, fast calculation speed, and high accuracy. It also makes full use of data resources and forms a standardized processing flow for water cut data, meeting the needs of automated production management in oil fields.

[0049] (5) It should be emphasized that the LSTM water cut prediction model constructed in this invention is not limited to data collected by the online measurement system for water cut in oil well production. Any measurement value related to time series can be predicted using the LSTM intelligent analysis module provided in this invention. Currently, distributed optical fiber acoustic sensor (DAS) technology is a key development direction for dynamic monitoring of oil wells, and it has broad application prospects in the monitoring and interpretation of oil well production and water cut. The acoustic data collected by optical fiber DAS technology is time series data. Therefore, the LSTM intelligent analysis module constructed in this invention can also be used for water cut measurement and prediction using optical fiber DAS technology. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the structure of the high-precision water content testing device for oil well production fluid based on intelligent modeling used in this invention;

[0051] Figure 2 A schematic diagram of the upper and lower swirl shapers;

[0052] Figure 3 This is a schematic diagram of the intermediate swirling radial differential pressure generator;

[0053] Figure 4 This is a schematic diagram of the technical measures of the data acquisition and LSTM intelligent analysis module of the present invention;

[0054] Figure 5This is a schematic diagram of the structure of a recurrent neural network (RNN);

[0055] Figure 6 This is a schematic diagram of the structure of a Long Short-Term Memory (LSTM) neural network.

[0056] Figure 7 This is a graph showing the real-time monitoring results of pressure difference and moisture content in the example;

[0057] Figure 8 This is a schematic diagram illustrating the relationship between the pressure difference ratio ZR and the moisture content in the example.

[0058] Figure 9 This is a comparison diagram of indoor experiments using a single rectifier and a dual rectifier in the embodiment.

[0059] Figure 10 This is a schematic diagram of the characteristic values ​​of the factors affecting water content as determined by principal component analysis in the examples.

[0060] Figure 11 This is a schematic diagram of the water content training results of the LSTM model in the example.

[0061] Figure 12 This is a schematic diagram of the training set error of the LSTM model in the embodiment;

[0062] Figure 13 This is a schematic diagram of the water content prediction results of the test set of the LSTM model in the embodiment;

[0063] Figure 14 This is a schematic diagram illustrating the results of using the LSTM model to predict future moisture content in the example.

[0064] In the diagram, 1 is the inlet section; 2 is the flow stabilizing pipe section; 3 is the upper connecting flange; 4-1 is the upper axial differential pressure chamber; 4-2 is the lower axial differential pressure chamber; 4-3 is the upper radial differential pressure chamber; 4-4 is the lower radial differential pressure chamber; 5 is the body sealing chamber; 6 is the lower connecting flange; 7-1 is the upper vortex shaper; 7-2 is the middle vortex radial differential pressure generator; 7-3 is the lower vortex shaper; 8-1 is the axial differential pressure bend; 8-2 is the radial differential pressure straight pipe; 9-1 is the inlet pressure sensor; 9-2 is the outlet temperature sensor; 9-3 is the radial differential pressure sensor; 9-4 is the axial differential pressure sensor; 10-1 is the data acquisition and LSTM intelligent analysis module; 10-2 is the wireless data transmission module; 10-3 is the inlet pressure signal transmission line; 10-4 is the outlet temperature signal transmission line; 10-5 is the axial differential pressure signal transmission line; 10-6 is the radial differential pressure signal transmission line. Detailed Implementation

[0065] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0066] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0067] To address the shortcomings of traditional intelligent modeling-based high-precision testing devices for water cut in oil wells, such as limited modeling parameters and low accuracy, this invention provides a method that utilizes machine learning algorithms for oilfield water cut interpretation, improving the accuracy of online water cut measurement data and predicting water cut in oil wells. Simultaneously, by incorporating the throttling friction within the testing device, it accurately interprets the production flow rate, achieving precise flow rate interpretation while simultaneously monitoring water cut. This solves the problem of measurement errors caused by the lack of simultaneous monitoring of water cut and flow rate in traditional techniques.

[0068] The high-precision water cut testing device for oil well production fluid based on intelligent modeling employs a dual-rectifier combined with a swirling radial differential pressure generator. The dual-rectifier approach overcomes the limitations of traditional rectifiers, such as narrow applicability and short flow pattern stability length. This allows for the formation of a stable annular flow with a water ring on the outside and an oil core in the center within both rectifiers. When oil and water swirl within the pipe, the viscosity difference between them generates swirling radial pressure differential and axial resistance pressure differential. Radial pressure differential refers to the pressure difference between the pipe wall and the center of the pipe at the same flow cross-section, which only occurs when the fluid rotates. Axial pressure differential refers to the pressure difference along the pipe axis, primarily reflecting the flow resistance at one end of the measuring device. Both pressure differentials are highly sensitive to flow rate and water cut. Therefore, the LSTM (Long Short-Term Memory) neural network intelligent analysis method is used to establish the relationship between the monitored values ​​and water cut, overcoming the shortcomings of traditional indoor experimental calibration models for determining flow rate and water cut, which suffer from low accuracy and narrow applicability.

[0069] This invention utilizes principal component analysis (PCA) and long short-term memory neural network (LSTM) for high-precision intelligent modeling to achieve short-term prediction of future water cut, establishing a standardized processing flow for water cut test data. It also provides a high-precision testing device for water cut in oil well production based on intelligent modeling. The system has a reasonable structure, overcoming the problems of narrow applicability and short flow pattern stability length of traditional rectifiers. The testing system embeds data acquisition and LSTM intelligent analysis modules, allowing for real-time updates of data and models, overcoming the shortcomings of low accuracy and narrow applicability of traditional indoor experimentally calibrated deterministic flow water cut models.

[0070] This invention achieves the optimization of structural parameters of a dual-rectifier oil well water cut testing system and the prediction of water cut in oil well production using LSTM intelligent modeling. The process is rigorous, practically applicable, and meets the needs of automated production management in oilfields. It is not limited to data collected by online water cut measurement systems; any time-series related measurements can be predicted using the LSTM intelligent analysis module provided by this invention.

[0071] The following are several specific embodiments of the application of the present invention.

[0072] Example 1

[0073] In a specific embodiment 1 of the present invention, the structural schematic diagram of the high-precision testing device for water cut in oil well production fluid based on intelligent modeling is shown below. Figure 1 As shown, it mainly includes the connection body part, the swirl shaping and differential pressure generation part, the test data monitoring part, and the data acquisition and intelligent analysis part. This involves 1. Inlet section, 2. Flow stabilizing pipe section, 3. Upper connecting flange, 4-1 Upper axial differential pressure chamber, 4-2 Lower axial differential pressure chamber, 4-3 Upper radial differential pressure chamber, 4-4 Lower radial differential pressure chamber, 5. Body sealing chamber, 6. Lower connecting flange, 7-1 Upper vortex shaper, 7-2 Middle vortex radial differential pressure generator, 7-3 Lower vortex shaper, 8-1 Axial differential pressure bend, 8-2 Radial differential pressure straight pipe, 9-1 Inlet pressure sensor, 9-2 Outlet temperature sensor, 9-3 Radial differential pressure sensor, 9-4 Axial differential pressure sensor, 10-1 Data acquisition and LSTM intelligent analysis module, 10-2 Wireless data transmission module, 10-3 Inlet pressure signal transmission line, 10-4 Outlet temperature signal transmission line, 10-5 Axial differential pressure signal transmission line, and 10-6 Radial differential pressure signal transmission line.

[0074] The connecting body includes: 1. Inlet section, 2. Flow stabilizing pipe section, 3. Upper connecting flange, 4-1. Upper axial differential pressure chamber, 4-2. Lower axial differential pressure chamber, 4-3. Upper radial differential pressure chamber, 4-4. Lower radial differential pressure chamber, 5. Body sealing chamber, and 6. Lower connecting flange. Inlet section 1 connects to flow stabilizing pipe section 2 to ensure a stable and steady flow rate of the incoming product liquid. Then, the upper connecting flange 3 sequentially connects to the upper axial differential pressure chamber 4-1, the upper radial differential pressure chamber 4-3, the body sealing chamber 5, the lower radial differential pressure chamber 4-4, and the lower axial differential pressure chamber 4-2 to form the main body. Finally, it connects to the lower connecting flange 6 and leads to the outlet.

[0075] The swirl shaping and differential pressure generation section includes: 7-1 upper swirl shaper, 7-2 middle swirl radial differential pressure generator, 7-3 lower swirl shaper, 8-1 axial differential pressure tapping bend, and 8-2 radial differential pressure tapping straight pipe. The upper swirl shaper 7-1, the middle swirl radial differential pressure generator 7-2, and the lower swirl shaper 7-3 are connected sequentially to form the core hardware of this test method. The measured axial and radial differential pressures are transmitted to the test data monitoring section through the axial differential pressure tapping bend 8-1 and the radial differential pressure tapping straight pipe 8-2.

[0076] The test data monitoring section includes: an inlet pressure sensor 9-1 and an outlet temperature sensor 9-2, which mainly collect the temperature and pressure of the product liquid, using traditional electronic temperature and pressure testing methods. The radial differential pressure sensor 9-3 mainly tests the radial differential pressure generated by the intermediate swirling radial differential pressure generator 7-2, while the axial differential pressure sensor 9-4 mainly tests the friction resistance along the flow path of the upper swirling shaper 7-1, the intermediate swirling radial differential pressure generator 7-2, and the lower swirling shaper 7-3.

[0077] The data acquisition and intelligent analysis section includes: a main module 10-1, which is primarily used to acquire data from the test data monitoring section and perform flow rate and water content testing and analysis using a built-in LSTM (Long Short-Term Memory) neural network algorithm; a wireless data transmission module 10-2, which transmits temperature, pressure, flow rate, and water content data remotely; and inlet pressure signal transmission line 10-3, outlet temperature signal transmission line 10-4, axial differential pressure signal transmission line 10-5, and radial differential pressure signal transmission line 10-6, which sequentially transmit data from the test data monitoring section to the main module.

[0078] The flow rate measurement method involves testing the radial differential pressure generated by the intermediate swirl radial differential pressure generator 7-2. The axial differential pressure sensor 9-4 primarily measures the friction resistance along the flow path of the upper swirl shaper 7-1, the intermediate swirl radial differential pressure generator 7-2, and the lower swirl shaper 7-3. This resistance signal is strong, resulting in high flow rate measurement accuracy. The relationships between axial pressure difference and flow rate are as follows:

[0079]

[0080] Where, ΔP z For axial pressure difference; α z ρ is the flow coefficient corresponding to the axial pressure difference; Q is the flow rate; A is the cross-sectional area of ​​the pipe; ρ is the density of the fluid.

[0081] The aforementioned dual-rectifier method based on in-pipe phase separation uses two cyclones connected in series to generate a phase separation section of sufficient length and stable morphology, so that an intermediate cyclone radial differential pressure generator can be placed in this section to carry out measurements. Studies have shown that the diameter of the formed in-pipe phase separation liquid column changes more smoothly in the axial direction than that of a single rectifier, and the optimal installation position of the intermediate cyclone radial differential pressure generator in this patent is given.

[0082] The structures of the upper and lower swirl shapers are shown below. Figure 2 .

[0083] The structure of the intermediate swirling radial differential pressure generator is shown below. Figure 3 .

[0084] Example 2

[0085] In a specific embodiment 2 of the present invention, the technical measures of the data acquisition and LSTM intelligent analysis module are mainly achieved through... Figure 4 This is achieved through a specific process.

[0086] Step 101: Data preprocessing, using PCA method to screen the main factors affecting water content.

[0087] Data collected using a high-precision water cut testing device for oil well production fluids based on intelligent modeling was used to form a time-series dataset for analyzing the main influencing factors of water cut. Factors related to water cut include radial pressure difference, axial pressure difference, pressure difference ratio (axial pressure difference / radial pressure difference, abbreviated as ZR), production fluid temperature, water cut factor variation, flow coefficient, measured flow rate, and structural parameters of the cyclone device (pipe diameter, hydrocyclone angle, hydrocyclone rotation angle, and pressure tapping method), etc. Among these, the selection of the structural parameters of the cyclone device is related to the two pressure difference values, and it is necessary to ensure that the pressure difference values ​​are within the reasonable operating range of the transmitter.

[0088] If there are more than 10 factors affecting water content, principal component analysis (PCA) can be used to screen out the 5-6 factors most relevant to water content. PCA is a dimensionality reduction statistical technique. The purpose of PCA is to analyze the importance of different influencing factors to water content, eliminate factors with less impact on water content, improve data selection accuracy, reduce data dimensionality, speed up calculation, and make the subsequently trained LSTM model more accurate.

[0089] Step 102: Max-Min normalization processing to form the feature vector of water content influencing factors.

[0090] After PCA analysis in step 101, the water content influencing factors are subjected to max-min normalization to establish a standard machine learning dataset. The aim is to eliminate the influence of different dimensions between indicators, making the data comparable and improving the model's prediction accuracy and convergence speed. The max-min normalization transformation formula is:

[0091]

[0092] Among them, X norm X represents the result after data normalization; X represents the original data; X max X represents the maximum value in the data; min This represents the minimum value in the data.

[0093] Construct a feature vector sample set using the normalized data. Assume X t Let Y be the feature vector of water-bearing influencing factors at time t, and let Y be the predicted water-bearing value at time t. t Each feature vector contains n features (n≤10), numbered F1-F2. n .

[0094] Step 103: The feature vectors are divided into training set and test set.

[0095] The sample set processed in step 102 is processed according to a selected time t. o Divide into, t o Time and t o The sample data before time step t is used as the training set. o The sample data after the specified time is used as the test set.

[0096] Step 104: Create a sliding window, build an LSTM water content prediction model, and train it.

[0097] Create a sliding window, set the window time step to Δt, and input the feature vector X for the first Δt time interval. t To predict the water content at time Δt+1, the time step is shifted by Δt each time to predict the water content at the next time Δt+1, until the entire training set is iterated.

[0098] The LSTM model uses a sample set consisting of an input time series and an output time series. The input time series is the feature vector X of water content influencing factors. t X t It contains n features F1-F n The output time series is the predicted water content Y at time t. t .

[0099] The training set after step 103 is input into the constructed LSTM model. The training set is processed through the input gate, forget gate and output gate to obtain the output of the first LSTM layer. The output vector of the first LSTM layer is used as the input vector of the second LSTM layer, and so on, continuously iterating. The output of each LSTM layer is the input of the next layer.

[0100] Step 105: Optimize training parameters and evaluate training results.

[0101] According to a preferred embodiment of the present invention, the LSTM network constructed in step 104 can be configured with multiple hidden layers, the number of neurons in the hidden layers ranging from 100 to 1000, the number of training cycles ranging from 500 to 5000, and the batch size ranging from 10 to 50. Based on the training effect, the parameters are continuously optimized within the set range.

[0102] To evaluate the accuracy of the LSTM model in water content prediction, this invention preferably uses the coefficient of determination (R²). 2 The evaluation metrics are: root mean square error (RMSE), mean absolute percentage error (MAPE), and three other metrics. The specific relationship is as follows:

[0103]

[0104]

[0105]

[0106] Among them, y i This is the actual value. The average of the actual values. These are the predicted values ​​from the LSTM model.

[0107] The quality of LSTM model training is judged based on evaluation metrics. When the metric value is less than a given error value, the model training is good and meets the requirements for liquid water content prediction, thus forming an LSTM water content prediction model. Conversely, the model performance is poor and needs to be retrained until it meets the requirements.

[0108] Step 106: Test and validate the LSTM water content prediction model.

[0109] The test set divided in step 103 is input into the LSTM water content prediction model trained and optimized in steps 104 and 105 for prediction.

[0110] After the prediction is completed, the predicted water content is output, and the output predicted value is inversely normalized.

[0111] Compare the predicted moisture content with the actual moisture content measured by sampling to analyze whether the model's prediction effect meets the on-site requirements.

[0112] Step 107, LSTM prediction of future short-term water content.

[0113] Prepare a set of data collected by the online measurement system for water cut in produced fluids of dual-rectifier oil wells, and process the data in steps 101 and 102 to form a feature vector of water cut influencing factors.

[0114] The feature vector is input into the LSTM water content prediction model trained, optimized, and tested in steps 104, 105, and 106, and the model outputs the predicted water content value in the short term.

[0115] In step 101,

[0116] The pipe diameter of the measuring device is used to measure the characteristic of the flow coefficient changing with flow rate at different flow velocities. The flow measurement range is 10m. 3 / d-100m 3 / d, the preferred pipe diameters are φ18, φ19 and φ20.

[0117] The tip angle of the hydrocyclone refers to the angle between the cyclone blades and the cross-section of the pipe. The smaller the angle, the stronger the fluid rotation, the greater the radial and axial differential pressure, and the more sensitive it is to water content. The preferred tip angles are 38° and 40°.

[0118] The rotation angle of the hydrocyclone refers to the rotation angle of the cyclone blades around the pipe axis. The larger the rotation angle, the longer the blades, the longer the forced flow distance and time of the fluid in the hydrocyclone, the more complete and intense the rotation, the greater the radial and axial differential pressure generated, and the more sensitive it is to water content. The rotation angle is preferably 400° and 500°.

[0119] The pressure tapping method of the hydrocyclone includes the location of the pressure tapping section and the shape and size of the pressure guiding column.

[0120] In step 104,

[0121] LSTM stands for Long Short-Term Memory Neural Network, a special type of Recurrent Neural Network (RNN). The structure of an RNN is as follows: Figure 5 As shown, X represents the input layer, H represents the hidden layer, O represents the output layer, U represents the weights from the input layer to the hidden layer, V represents the weights from the hidden layer to the output layer, and W represents the weights from the hidden layer to the hidden layer. RNNs have advantages in solving time series problems because they can feed information from the previous state into the prediction of the next state, making the model output simultaneously influenced by the current input and all previous inputs. However, due to their simple structure, RNNs are prone to gradient vanishing or gradient exploding during training, meaning the weight coefficients between hidden layers tend to 0 or infinity.

[0122] LSTM improves upon RNN, outperforming it in both accuracy and training speed. Therefore, LSTM is more suitable for long-term or short-term time series data prediction in oil and gas field development. Compared to RNN, LSTM adds input gates, output gates, and forget gates. These gates allow the model processor to selectively retain previously valid information throughout the LSTM network, as shown in the structure... Figure 6 As shown. In this invention, the LSTM network processes and analyzes the input X at time t. t Long-term hidden C t-1 and short-term hidden H t-1 To generate output Y t C t-1 H contains information about the time steps prior to time t. t-1 Including information from the previous time step, X t and H t-1 Processed by a fully connected layer FC, where g t f t i t o t They are respectively:

[0123]

[0124]

[0125]

[0126]

[0127] In this invention, f is preferably a nonlinear activation function ReLU; σ is an activation function Sigmoid. t i t o t Control the forget gate, input gate, and output gate respectively; W xg W xf W xi W xo To process input X t The weight matrix, W hg W hf W hi W ho To handle short-term hidden H t-1 The weight matrix, b g b f b i b o This is a bias term.

[0128] LSTM models can achieve end-to-end prediction, including single-factor prediction of a single indicator, multi-factor prediction of a single indicator, and multi-factor prediction of multiple indicators. This invention utilizes multi-factor prediction of a single indicator, processing time-series data of multiple factors collected in chronological order, capturing the interrelationships between them and their trends over time, thereby using an LSTM model to predict the required indicator data.

[0129] Example 3

[0130] In a specific embodiment 3 of the present invention, a high-precision testing device for water cut in well production fluids based on intelligent modeling was used to collect data from well A from 0:00:00 on June 1, 2020 to 13:00:00 on June 8, 2020, with a sampling interval of 10 seconds. Due to the large amount of data and long calculation time, this embodiment performs data aggregation processing to extend the sampling interval to 30 minutes, resulting in a total of 316 sets of data. The water cut in production fluids of well A is then predicted using the technical method of the present invention.

[0131] The technical measures for data acquisition and LSTM intelligent analysis modules are mainly through... Figure 4 This is achieved through a specific process.

[0132] Step 101: Data preprocessing, using PCA method to screen the main factors affecting water content.

[0133] Data collected using a high-precision water cut testing device for oil well production fluids based on intelligent modeling were used to form a time-series dataset to analyze the main influencing factors of water cut. The water cut influencing factors collected from Well A included 13 factors: radial pressure difference, axial pressure difference, pressure difference ratio ZR, production fluid temperature, water cut factor variation, flow coefficient, measured flow rate, viscosity, gas-oil ratio, measuring device pipe diameter, hydrocyclone angle, hydrocyclone rotation angle, and pressure tapping method. Figure 7 The trends of radial and axial pressure differences with water content are shown, and it can be seen that the characteristics of these two pressure differences with water content are consistent. Figure 8 The water content f is shown. w It exhibits a linear correlation with the pressure difference ratio ZR, as shown in the following formula, with a linear correlation coefficient of 0.8653.

[0134] f w = -0.181·ZR + 18.883

[0135] In the embodiment, the indoor experimental comparison diagram of single rectifier and dual rectifier is shown in the figure below. Figure 9As shown. The preferred pipe diameter for the online water cut measurement device in a dual-rectifier oil well is φ20, the preferred hydrocyclone angle is 45°, and the preferred hydrocyclone rotation angle is 500°. In the pressure tapping method, when axial pressure tapping points are arranged, the first pressure tapping position is located between 2D and 5D upstream of the hydrocyclone, and the second pressure tapping position is located between 5D and 10D upstream of the hydrocyclone. The radial pressure tapping surface is arranged between 1D and 5D downstream of the hydrocyclone, and the second axial pressure tapping position can coincide with the pressure tapping on the radial pressure tapping wall.

[0136] Principal component analysis (PCA) is used to screen out the factors most relevant to water content. The purpose of PCA is to analyze the importance of different influencing factors to water content, eliminate factors with less influence on water content, improve the accuracy of data selection, reduce data dimensionality, speed up calculation, and make the LSTM model trained subsequently more accurate. Figure 10 The data shows the characteristic values ​​of different influencing factors on water cut after PCA processing. It can be seen that the characteristic values ​​of the six factors, namely the pressure difference ratio ZR, axial pressure difference, radial pressure difference, measured flow rate, flow coefficient, and water cut factor change, are relatively large and belong to the first six principal components. The sum of the characteristic values ​​of these six principal components accounts for 82.6% of the total characteristic values, indicating that the contribution rate of these six major influencing factors to the water cut data is as high as 82.6%.

[0137] Step 102: Max-Min normalization processing to form the feature vector of water content influencing factors.

[0138] After PCA analysis in step 101, the water content influencing factors are subjected to max-min normalization to establish a standard machine learning dataset. The aim is to eliminate the influence of different dimensions between indicators, making the data comparable and improving the model's prediction accuracy and convergence speed. The max-min normalization transformation formula is:

[0139]

[0140] Among them, X norm X represents the result after data normalization; X represents the original data; X max X represents the maximum value in the data; min This represents the minimum value in the data.

[0141] Construct a feature vector sample set using the normalized data. Assume X t Let Y be the feature vector of water-bearing influencing factors at time t, and let Y be the predicted water-bearing value at time t. t In the embodiment, each feature vector contains 6 features, numbered F1-F6, which respectively represent the pressure difference ratio ZR, axial pressure difference, radial pressure difference, measured flow rate, flow coefficient and water content factor change.

[0142] Step 103: The feature vectors are divided into training set and test set.

[0143] The sample set processed in step 102 is processed according to a selected time t. o In the example, t is divided into segments. o The time was set as 9:00:00 on June 6th. o Time and t o The sample data before time step t is used as the training set. o The sample data after time step is used as the test set and the data for predicting water content in the short term. The number of samples in the training set is 267, the number of samples in the test set is 30, and the number of samples used for predicting water content in the short term is 10.

[0144] Step 104: Create a sliding window, build an LSTM water content prediction model, and train it.

[0145] Create a sliding window, set the window step size Δt to 10 (i.e., 5 hours of data), and input the feature vector X for the first Δt time period. t To predict the water content at time Δt+1, the time step is shifted by Δt each time to predict the water content at the next time Δt+1, until the entire training set is iterated.

[0146] The LSTM model uses a sample set consisting of an input time series and an output time series. The input time series is the feature vector X of water content influencing factors. t X t It contains 6 features F1-F6, and the output time series is the predicted water content Y at time t. t .

[0147] The training set after step 103 is input into the constructed LSTM model. The training set is processed through the input gate, forget gate and output gate to obtain the output of the first LSTM layer. The output vector of the first LSTM layer is used as the input vector of the second LSTM layer, and so on, continuously iterating. The output of each LSTM layer is the input of the next layer.

[0148] Step 105: Optimize training parameters and evaluate training results.

[0149] According to a preferred embodiment of the present invention, the LSTM network is configured with two hidden layers, the first hidden layer having 256 neurons and the second hidden layer having 512 neurons, the number of training loops being 500, and the batch size being 10.

[0150] To evaluate the accuracy of the LSTM model in water content prediction, this invention preferably uses the coefficient of determination (R²). 2 The evaluation metrics are: root mean square error (RMSE), mean absolute percentage error (MAPE), and three other metrics. The specific relationship is as follows:

[0151]

[0152]

[0153]

[0154] Among them, y i This is the actual value. The average of the actual values. These are the predicted values ​​from the LSTM model.

[0155] The quality of LSTM model training is judged based on evaluation metrics. When the metric value is less than a given error value (error), the model training is considered good and meets the requirements for liquid water content prediction, thus forming an LSTM water content prediction model. Conversely, if the metric value is greater than a given error value (error), the model performance is poor and needs to be retrained until it meets the requirements. In this example, the error value (error) is set to 0.3.

[0156] Step 106: Test and validate the LSTM water content prediction model.

[0157] The test set divided in step 103 is input into the LSTM water content prediction model trained and optimized in steps 104 and 105 for prediction.

[0158] After the prediction is completed, the predicted water content is output, and the output predicted value is inversely normalized.

[0159] Compare the predicted moisture content with the actual moisture content measured by sampling to analyze whether the model's prediction effect meets the on-site requirements.

[0160] Step 107, LSTM prediction of future short-term water content.

[0161] Steps 101 and 102 are performed on the 10 sets of sample data used for short-term water content prediction in step 103 to form a feature vector of water content influencing factors.

[0162] The feature vector is input into the LSTM water content prediction model trained, optimized, and tested in steps 104, 105, and 106, and the model outputs the predicted water content value in the short term.

[0163] The training results of the LSTM model with water content in the training set are as follows: Figure 11 As shown, the training results of the LTSM model accurately grasped the trend of water cut change in well A, with an error of [missing information]. Figure 12 The results show that all different error evaluation metrics on the training set are within the set error range, meeting the accuracy requirements for water content prediction. The water content prediction results on the test set of the LSTM model are as follows: Figure 13As shown in Table 1, the statistical results have an average relative error of 0.78%. The PCA multi-factor analysis method and LSTM water cut prediction model established in this invention can predict the water cut changes in well A with high accuracy, demonstrating the effectiveness and significant advantages of the proposed method in solving the problem of oil well water cut prediction. Furthermore, using current sample data, predictions were made for the water cut over the next 10 hours, and the prediction results are shown below. Figure 14 As shown, the method proposed in this invention facilitates timely understanding of water content changes that may occur in the near future, enabling targeted implementation and control of on-site measures and guiding on-site production work.

[0164] Table 1. Water content prediction results of the LSTM model on the test set.

[0165]

[0166]

[0167] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. 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.

[0168] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. A high-precision testing device for water cut in oil well production fluid based on intelligent modeling, characterized in that, This high-precision water cut testing device for oil well produced fluid based on intelligent modeling includes a connecting body, a swirl shaping and differential pressure generation section, a test data monitoring section, and a data acquisition and intelligent analysis section. The swirl shaping and differential pressure generation section is connected to the connecting body, which ensures a stable and consistent produced fluid flow rate. This section swirls and shapes the produced fluid, generating radial and axial differential pressures. The test data monitoring section is connected to the swirl shaping and differential pressure generation section, collecting the radial differential pressure and friction resistance of the produced fluid and transmitting them to the data acquisition and intelligent analysis section. Based on the received data, the data acquisition and intelligent analysis section uses an LSTM (Long Short-Term Memory) neural network algorithm to perform flow rate and water cut testing and analysis.

2. The high-precision testing device for water cut in oil well production based on intelligent modeling according to claim 1, characterized in that, The main body of the connection includes an inlet section, a flow stabilizing pipe section, an upper connecting flange, an upper axial differential pressure chamber, a lower axial differential pressure chamber, an upper radial differential pressure chamber, a lower radial differential pressure chamber, a main body sealing chamber, and a lower connecting flange. The inlet section is connected to the flow stabilizing pipe section to ensure a stable and steady flow rate of the incoming product liquid. The upper connecting flange is connected to the flow stabilizing pipe section. The upper connecting flange is sequentially connected to the upper axial differential pressure chamber, the upper radial differential pressure chamber, the main body sealing chamber, the lower radial differential pressure chamber, and the lower axial differential pressure chamber to form the main body of the main body. Then, it is connected to the lower connecting flange and leads to the outlet.

3. The high-precision testing device for water cut in oil well production based on intelligent modeling according to claim 1, characterized in that, The swirl shaping and differential pressure generation section includes an upper swirl shaper, a middle swirl radial differential pressure generator, a lower swirl shaper, an axial differential pressure tapping bend, and a radial differential pressure tapping straight pipe. The upper swirl shaper, the middle swirl radial differential pressure generator, and the lower swirl shaper are connected in sequence to form a stable annular flow with a water ring on the outside and an oil core in the center. The measured axial and radial differential pressures are transmitted to the test data monitoring section through the axial differential pressure tapping bend and the radial differential pressure tapping straight pipe.

4. The high-precision testing device for water cut in oil well production based on intelligent modeling according to claim 3, characterized in that, The test data monitoring section includes an inlet pressure sensor, an outlet temperature sensor, a radial differential pressure sensor, and an axial differential pressure sensor. The inlet pressure sensor collects the pressure data of the product liquid, the outlet temperature sensor collects the temperature data of the product liquid, the radial differential pressure sensor collects the radial differential pressure generated by the intermediate swirling radial differential pressure generator, and the axial differential pressure sensor collects the friction resistance of the upper swirling shaper, the intermediate swirling radial differential pressure generator, and the lower swirling shaper.

5. The high-precision testing device for water cut in oil well production based on intelligent modeling according to claim 1, characterized in that, The data acquisition and intelligent analysis section receives data transmitted from the test data monitoring section and calculates the relationship between axial pressure difference and flow rate as follows: Where, ΔP z For axial pressure difference; α z ρ is the flow coefficient corresponding to the axial pressure difference; Q is the flow rate; A is the cross-sectional area of ​​the pipe; ρ is the density of the fluid.

6. A high-precision testing and interpretation method for water cut in oil well production fluid based on intelligent modeling, characterized in that, The method for high-precision testing and interpretation of water content in oil well production fluid based on intelligent modeling employs the high-precision testing device for water content in oil well production fluid based on intelligent modeling as described in claim 1, comprising: Step 1: Perform data preprocessing and use PCA method to screen the main factors affecting water content; Step 2: Perform max-min normalization to form the feature vector of water content influencing factors; Step 3: Divide the feature vectors into a training set and a test set; Step 4: Create a sliding window, build an LSTM water content prediction model, and train it; Step 5: Optimize training parameters and evaluate training results; Step 6: Test and validate the LSTM water content prediction model; Step 7: Perform LSTM short-term prediction of future water content.

7. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 6, characterized in that, In step 1, a time series dataset is formed based on the data collected by the high-precision water cut testing device for oil well production fluid based on intelligent modeling, and the main influencing factors of water cut are analyzed. Factors related to water cut include radial pressure difference, axial pressure difference, pressure difference ratio (axial pressure difference / radial pressure difference ZR), production fluid temperature, water cut factor variation, flow coefficient, measured flow rate, and structural parameters of the cyclone device, including the pipe diameter of the measuring device, the hydrocyclone angle, the hydrocyclone rotation angle, and the pressure tapping method.

8. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 7, characterized in that, In step 1, if there are many factors affecting water content, more than 10, then principal component analysis (PCA) is used to screen out the 5-6 factors most relevant to water content.

9. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 6, characterized in that, In step 2, the water content influencing factors after PCA analysis in step 1 are subjected to maximum-minimum normalization to establish a standard machine learning dataset. The maximum-minimum normalization transformation formula is: Among them, X norm X represents the result after data normalization; X represents the original data; X max X represents the maximum value in the data; min This represents the minimum value in the data; Construct a feature vector sample set using the normalized data; assume X t Let Y be the feature vector of water-bearing influencing factors at time t, and let Y be the predicted water-bearing value at time t. t Each feature vector contains n features, n≤10, numbered F1-F1. n .

10. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 6, characterized in that, In step 3, the sample set processed in step 2 is processed according to a selected time t. o Divide into, t o Time and t o The sample data before time step t is used as the training set. o The sample data after the specified time is used as the test set.

11. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 6, characterized in that, In step 4, create a sliding window, set the window time step to Δt, and input the feature vector X for the first Δt time interval. t To predict the water content at time Δt+1, the time step is shifted by Δt each time to predict the water content at the next time Δt+1, until the entire training set is iterated.

12. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 11, characterized in that, In step 4, the sample set used by the LSTM model consists of the input time series and the output time series. The input time series is the feature vector X of water content influencing factors. t X t It contains n features F1-F n The output time series is the predicted water content Y at time t. t The training set after step 3 is input into the constructed LSTM model. The training set is processed through the input gate, forget gate and output gate to obtain the output of the first LSTM layer. The output vector of the first LSTM layer is used as the input vector of the second LSTM layer, and so on, iterating continuously. The output of each LSTM layer is the input of the next layer.

13. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 6, characterized in that, In step 5, to evaluate the accuracy of the LSTM model in water content prediction, the optimal coefficient of determination R is selected. 2 The three evaluation indicators are: Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE); the specific relationship is as follows: Among them, y i This is the actual value. The average of the actual values. These are the predicted values ​​from the LSTM model; The quality of LSTM model training is judged by evaluation metrics. When the metric value is less than the given error value, the model training is good and meets the requirements for liquid water content prediction, and an LSTM water content prediction model can be formed. Otherwise, the model performance is poor and needs to be retrained until it meets the requirements.

14. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 6, characterized in that, In step 6, the test set divided in step 3 is input into the LSTM water content prediction model trained and optimized in steps 4 and 5 for prediction; after the prediction is completed, the predicted water content value is output, and the output predicted value is inversely normalized; the predicted water content value is compared with the actual sampled water content value to analyze whether the model prediction effect meets the field requirements.

15. The method for high-precision testing and interpretation of water cut in oil well production fluid based on intelligent modeling according to claim 6, characterized in that, In step 7, prepare a set of data collected through high-precision testing of water content in oil well production fluid based on intelligent modeling, process the data according to steps 1 and 2, and form a feature vector of water content influencing factors. The feature vector is input into the LSTM water content prediction model trained, optimized, and tested in steps 4, 5, and 6, and the model outputs the predicted water content value in the short term.

Images