A crude oil water cut measurement device and method based on a GRU neural network model

Abstract

This invention relates to the field of fluid measurement technology, specifically a crude oil water content measurement device and method based on a GRU neural network model. The device includes a longitudinal five-electrode array sensor, a front-end signal processing and measurement circuit module, a microprocessor control module, and a temperature monitoring module. The longitudinal five-electrode array sensor is connected to the front-end signal processing and measurement circuit module, which is connected to the microprocessor control module. The microprocessor control module is connected to the temperature monitoring module, which is connected to the longitudinal five-electrode array sensor. This invention uses a microprocessor control module as the main controller and combines it with various conversion circuits to complete the measurement of signal excitation, phase content, and related flow velocities. It features low energy consumption, strong anti-interference capability, and a high degree of automation.

Description

Technical Field

[0001] This invention relates to the field of fluid measurement technology, and in particular to a crude oil water content measurement device and method based on a GRU neural network model. Background Technology

[0002] Crude oil water content is a crucial data point for crude oil production and storage, and a key parameter for studying oilfield development. Detection of crude oil water content is essential during production and storage. However, crude oil water content is a complex parameter susceptible to multiple factors with uncertain relationships, making it difficult to characterize using a unified mathematical model or formula. This limits the effectiveness of single algorithms, hindering the accuracy of existing online instruments in sampling and limiting the development of crude oil water content measurement methods. As most oilfields in my country enter the late stages of production, the water content of the produced crude oil is high, and the variation in water content at wellhead sampling is significant, greatly increasing the difficulty of manual sampling. While manual sampling offers high accuracy, the long sampling cycle makes it impossible to monitor changes in crude oil water content in real time to meet the needs of production, storage, and transportation, let alone automated production. Online instruments can achieve real-time measurement, but under the influence of external factors such as temperature and salinity, they struggle to meet the reliability and stability requirements of actual oilfield production. Therefore, researching crude oil water content measurement methods is crucial for improving the accuracy of crude oil water content detection instruments and the level of automated production. Summary of the Invention

[0003] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a crude oil water content measurement device and method based on the GRU neural network model, which solves the technical problems of long measurement cycle and low accuracy in the prior art.

[0004] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0005] On one hand, the present invention provides a crude oil water content measurement device based on a GRU neural network model, comprising a longitudinal five-electrode array sensor, a front-end signal processing and measurement circuit module, a microprocessor control module, and a temperature monitoring module; the longitudinal five-electrode array sensor is connected to the front-end signal processing and measurement circuit module, the front-end signal processing and measurement circuit module is connected to the microprocessor control module; the microprocessor control module is connected to the temperature monitoring module, and the temperature monitoring module is connected to the longitudinal five-electrode array sensor.

[0006] Furthermore, the longitudinal five-electrode array sensor includes a cylinder, and a pair of excitation electrodes, a pair of correlation velocity measurement electrodes, and an independent phase content measurement electrode sleeved on the cylinder; the excitation electrodes are disposed at both ends of the cylinder, the correlation velocity measurement electrodes are disposed on the cylinder inside the excitation electrodes, and the phase content measurement electrode is disposed in the middle of the cylinder; the excitation electrodes are connected to the front-end signal processing and measurement circuit module; the correlation velocity measurement electrodes are connected to the front-end signal processing and measurement circuit module; and the phase content measurement electrode is connected to both the front-end signal processing and measurement circuit module and the temperature monitoring module.

[0007] Furthermore, the excitation electrode, the related flow rate measuring electrode, and the phase content measuring electrode are all made of stainless steel rings.

[0008] Furthermore, the front-end signal processing and measurement circuit module includes a signal generation circuit, a differential amplifier circuit, a limiting circuit, and an ADC sampling circuit; the signal generation circuit is connected to the excitation electrode via the differential amplifier circuit; the differential amplifier circuit is connected to the limiting circuit, the limiting circuit is connected to the ADC sampling circuit, and the ADC sampling circuit is connected to the microprocessor control module.

[0009] Furthermore, the microprocessor control module is a single-chip microcomputer, and the microprocessor control module is connected to the ADC sampling circuit. The other end of the ADC sampling circuit is connected to the limiting circuit. The microprocessor control module uses the cross-correlation velocity algorithm based on Fourier transform from the DSP library and the water content algorithm based on Fourier transform from the DSP library to complete the measurement of the water content of crude oil.

[0010] Furthermore, the temperature monitoring module includes an analog output temperature sensor, a constant current source circuit, and an OLED screen; the analog output temperature sensor is connected to the constant current source circuit, and the OLED screen is connected to the microprocessor control module.

[0011] On the other hand, the present invention provides a method for measuring the water content of crude oil based on a GRU neural network model, comprising the following steps:

[0012] S1. Apply an AC current with an amplitude of 5V and an output frequency of 18KHz to the excitation electrode of the longitudinal five-electrode array sensor as the excitation power supply for the longitudinal five-electrode array sensor.

[0013] S2. When the fluid flows through the longitudinal five-electrode array sensor, the electric field generated by a pair of excitation electrodes produces a weak signal on the phase content measurement electrode. After being processed by the front-end signal processing measurement circuit module, the signal is sampled by the ADC circuit to obtain an array consisting of 4096 voltage elements. The array is then subjected to a fast Fourier transform based on the DSP library with 4096 points using the GRU neural network model. The crude oil water content information is concentrated in a certain fixed frequency band. The phase content measurement can be completed by calibrating based on the water content data.

[0014] S3. A microprocessor control module is used to calculate the cross-correlation flow rate algorithm based on FFT acceleration and the moisture content algorithm based on FFT in the DSP library, and the results are output to the OLED screen of the temperature monitoring module.

[0015] Furthermore, in step S2: the GRU neural network model is specifically as follows:

[0016] S21. Training data: In the GRU neural network model, 4096 voltage elements collected by the ADC sampling circuit are used as the training set for training; the convolutional layer is set with a large convolutional kernel of 125×1. The GRU neural network model uses overlapping pooling with a pooling kernel length greater than the stride. The pooling kernel sizes are 5×1 and 3×1, and the stride of the two pooling layers is set to 2 to obtain a preliminary model.

[0017] S22, Model Training: The essence of measuring moisture content data using a GRU neural network model is to extract key features from the corresponding samples for classification, and to use a logarithmic loss function to measure the difference between the predicted value and the true input value; the final loss function L of the GRU neural network model is defined, and the gradient of the hidden layer is:

[0018]

[0019] The gradients corresponding to different parameter matrices can be calculated:

[0020]

[0021]

[0022]

[0023] S23. Model Performance Evaluation: The GRU neural network model performs adaptive crude oil water content measurement on the collected phase content signals. First, the divided dataset is input into the GRU neural network model for training. The parameters are adjusted through the adaptive moment estimation algorithm to make the GRU neural network model converge. After obtaining the optimal training results, the model is saved. Then, crude oil water content is measured on the test set, and the diagnostic results are analyzed to verify the effectiveness of the GRU neural network model.

[0024] Further, in step S3: the output result value is displayed on the OLED screen of the temperature monitoring module, specifically: the temperature value detected by the analog output temperature sensor is sent to the microprocessor control module for processing through the ADC sampling circuit and the constant current source circuit, and the result is displayed on the OLED screen.

[0025] This invention provides a crude oil water content measurement device and method based on a GRU neural network model. Compared with the prior art, the advantages of this invention are as follows:

[0026] 1. The crude oil water cut measurement method based on the GRU neural network model of this invention has good data processing performance. The GRU neural network model algorithm is used to predict the phase content of oil-water two-phase flow in a vertical riser pipe. Experimental data under various operating conditions were processed, and feature quantities were extracted in both the time and frequency domains. Then, the GRU neural network model was used for analysis, achieving high-precision water cut prediction results.

[0027] 2. This invention uses a microcontroller as the main controller and combines it with various conversion circuits to complete the measurement of signal excitation, phase content, and related flow velocities. It features low energy consumption, strong anti-interference capability, and a high degree of automation.

[0028] 3. This invention uses a stainless steel metal ring and, considering the ionization problem, uses an 18kHz sine wave signal as the excitation signal, which effectively prevents corrosion of the metal ring and improves the service life of the sensor. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the structure of the conductivity-type longitudinal multipole array sensor provided in an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of the structure and functional modules of the conductivity-type longitudinal multipole array sensor provided in an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of the signal excitation circuit for a conductivity-type longitudinal multipole array sensor provided in an embodiment of the present invention.

[0032] Figure 4The schematic diagram of the signal processing circuit of the conductivity-type longitudinal multipole array sensor provided in the embodiment of the present invention.

[0033] In the figure: 1. Cylinder; 2. Excitation electrode; 3. Correlation velocity measurement electrode; 4. Phase content measurement electrode. Detailed Implementation

[0034] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below through specific embodiments.

[0035] This invention provides a crude oil water content measurement device and method based on a GRU neural network model. It utilizes an embedded system combined with five metal rings deployed on a pipeline as detection sensors to improve the measurement accuracy of crude oil water content. The five metal rings include: a pair of excitation electrodes, a pair of relative flow velocity measuring electrodes, and an independent phase content measuring electrode. Simultaneously, a temperature detection system is designed to correct for the influence of temperature on conductivity.

[0036] Example 1:

[0037] Reference Figures 1 to 3 This invention provides a crude oil water content measurement device based on a GRU neural network model. The device includes a longitudinal five-electrode array sensor, a front-end signal processing and measurement circuit module, a microprocessor control module, and a temperature monitoring module. The longitudinal five-electrode array sensor is connected to the front-end signal processing and measurement circuit module, which is connected to the microprocessor control module. The microprocessor control module is connected to the temperature monitoring module, which is connected to the longitudinal five-electrode array sensor.

[0038] Specifically, the longitudinal five-electrode array sensor includes a cylinder 1, a pair of excitation electrodes 2, a pair of correlation velocity measuring electrodes 3, and an independent phase content measuring electrode 4, all mounted on the cylinder 1. The excitation electrodes 2 are located at both ends of the cylinder 1, the correlation velocity measuring electrodes 3 are located on the cylinder 1 inside the excitation electrodes 2, and the phase content measuring electrode 4 is located in the middle of the cylinder 1. The excitation electrodes 2 are connected to a front-end signal processing and measurement circuit module. The correlation velocity measuring electrodes 3 are also connected to the front-end signal processing and measurement circuit module. The phase content measuring electrode 4 is connected to both the front-end signal processing and measurement circuit module and a temperature monitoring module. When water-containing crude oil flows through this longitudinal five-electrode array sensor, the equivalent impedance of the sensor depends on the water content of the crude oil. Therefore, when an excitation voltage signal is applied between the excitation electrodes of the sensor, a voltage signal is also output between the correlation velocity measuring electrodes, and the amplitude of this voltage signal is related to the water content.

[0039] The excitation electrode, the correlation velocity measurement electrode, and the phase content measurement electrode are all made of stainless steel rings.

[0040] Specifically, the front-end signal processing and measurement circuit module includes a signal generation circuit and a signal detection circuit. The signal detection circuit includes a differential amplifier circuit, a limiting circuit, and an ADC sampling circuit; the ADC sampling circuit is model AD7906. The signal generation circuit is connected to the excitation electrode via the differential amplifier circuit; the differential amplifier circuit is connected to the limiting circuit, the limiting circuit is connected to the ADC sampling circuit, and the ADC sampling circuit is connected to the microprocessor control module.

[0041] The signal generation circuit consists of a direct digital frequency synthesis to generate a high-frequency waveform generator. It uses a direct digital fitting (DDS) chip to generate a high-frequency sinusoidal voltage source, and then uses a voltage-controlled current source (VCCS) to convert the sinusoidal voltage signal into a sinusoidal current excitation source signal.

[0042] The signal detection circuit includes a differential amplifier circuit, a limiting circuit, and an ADC sampling circuit, which are used in combination to amplify the input signal, eliminate high-frequency noise generated by the DDS chip, and acquire data.

[0043] Those skilled in the art will understand that other similar demodulation methods can also implement the present invention. For example, the PXI-4472 data communication acquisition card, etc.

[0044] Specifically, the microprocessor control module is a single-chip microcomputer, specifically an STM32F107. The microprocessor control module is connected to the ADC sampling circuit, and the other end of the ADC sampling circuit is connected to the limiting circuit. The microprocessor control module utilizes the Fourier transform-based cross-correlation velocity algorithm from the DSP library and the Fourier transform-based water content algorithm from the DSP library to measure the water content of crude oil.

[0045] Specifically, the temperature monitoring module includes an analog output temperature sensor, a constant current source circuit, and an OLED screen; the analog output temperature sensor is connected to the constant current source circuit, and the OLED screen is connected to the microprocessor control module. The constant current source circuit is model XTR105.

[0046] When the measuring device is working, an 18kHz sine wave signal is generated as the excitation signal through direct digital frequency synthesis. The microprocessor control module generates an interrupt signal at regular intervals to control the signal generation circuit to output a strictly symmetrical excitation voltage signal to the longitudinal five-electrode array sensor. The measurement signal output by the relevant flow velocity measuring electrodes of the longitudinal five-electrode array sensor is directly input to the signal detection circuit for effective value processing, and then fed back to the microprocessor control module for data fitting. The signal generation circuit can discharge at regular intervals according to the signal provided by the microprocessor control module to eliminate charge accumulation and improve measurement accuracy.

[0047] Example 2:

[0048] This invention provides a method for measuring the water content of crude oil based on a GRU neural network model, comprising the following steps:

[0049] S1. Apply an AC current with an amplitude of 5V and an output frequency of 18KHz to the excitation electrode of the longitudinal five-electrode array sensor as the excitation power supply for the longitudinal five-electrode array sensor.

[0050] S2. When fluid flows through the longitudinal five-electrode array sensor, the electric field generated by the pair of excitation electrodes will undergo abnormally complex distortion and deformation due to the randomly distributed oil bubbles inside the two-phase fluid. This results in the pair of related velocity measurement electrodes detecting velocity-related signals. These signals have the same frequency and amplitude, only different phases. The velocity relationship can be measured using the phase relationship. Specifically, when fluid flows through the longitudinal five-electrode array sensor, the electric field generated by the pair of excitation electrodes produces a weak signal at the phase content measurement electrode. After being processed by the front-end signal processing measurement circuit module, the signal is sampled by the ADC circuit to obtain an array consisting of 4096 voltage elements. A GRU neural network model is used to perform a fast Fourier transform on the array with 4096 points based on the DSP library. The crude oil water content information is concentrated in a certain fixed frequency range. The phase content can be measured by calibrating based on the water content data.

[0051] S3. A microprocessor control module is used to calculate the cross-correlation flow rate algorithm based on FFT acceleration and the moisture content algorithm based on FFT in the DSP library. The analog output temperature value detected by the temperature sensor is sent to the STM32F107 microcontroller for processing through the ADC sampling circuit and the XTR105 constant current source circuit, and the result is output to the OLED screen.

[0052] The signal detected by the phase fill rate measuring electrode contains historical data on phase fill rate, which is a curve showing the relationship between voltage and phase fill rate. Since data processing takes a long time, a neural network is needed to process the data. This invention uses a GRU neural network to solve this problem.

[0053] First, all relevant historical data needs to be collected. This invention divides the historical data into two parts according to date and time: the most recent 4096 historical data points and the last 4096 historical data points. To facilitate processing by the GRU neural network, this invention uses the first 4096 data points as the training set and the last 4096 data points as the test set. Since the last 4096 data points are not representative, a nearest neighbor (Neural Network) is used for modeling. The first 4096 historical samples contain most of the variables related to the consistency ratio. This invention uses the Gated Recurrent Unit (GRU) in the GRU neural network model to process these samples.

[0054] Since historical data has already been collected, training on the data from the last month is sufficient to achieve good results. This is a GRU neural network model composed of an LSTM network. The main steps are as follows:

[0055] 1. Training data: The original voltage data is collected as the training set. This part is then used to train the GRU neural network to obtain a preliminary model.

[0056] 2. Initialization: The initial model is then input into a neural network for training, and parameters are adjusted to obtain optimal performance. If using a GRU neural network, this process can be treated as a convolutional neural network (CNN). Finally, the data processing for the content ratio is completed.

[0057] Specifically, in step S2: the GRU neural network model is as follows:

[0058] S21. Training Data: In the GRU neural network model, 4096 voltage elements collected by the ADC sampling circuit were used as the training set for training. A large 125×1 convolutional kernel was designed for the convolutional layers to accelerate the model's convergence speed. To prevent the omission of key local features during the convolution process, the GRU neural network model used overlapping pooling with a kernel length greater than the stride. The kernel sizes were 5×1 and 3×1, respectively, and the stride of the two pooling layers was set to 2, resulting in a preliminary model.

[0059] S22. Model Training: The purpose of training the GRU neural network model is to continuously update the weight information and fit the data distribution of the input samples. The essence of measuring moisture content data using the GRU neural network model is to extract key features of the corresponding samples and classify them. Therefore, a logarithmic loss function is used to measure the difference between the predicted value and the true input value. Defining the final loss function L of the model, the gradient of the hidden layer is:

[0060]

[0061] The gradients corresponding to different parameter matrices can be calculated:

[0062]

[0063]

[0064]

[0065] S23. Model Performance Evaluation: The GRU method adaptively measures crude oil water content using the collected phase content signals. First, the pre-divided dataset is input into the model for training. The parameters are adjusted using the Adaptive Moment Estimation Algorithm (Adam) to achieve model convergence. The model is saved after obtaining the optimal training results. Then, crude oil water content is measured on the test set, and the diagnostic results are analyzed to verify the effectiveness of the GRU method. The GRU method is built on the TensorFlow deep learning framework developed by Google, version 2.1.0. The experimental model was developed using Python 3.6 and deployed on an Intel I7-6900K CPU computer.

[0066] Specifically, in step S3: the output result value is displayed on the OLED screen of the temperature monitoring module. Specifically, the temperature value detected by the analog output temperature sensor is sent to the microprocessor control module for processing through the ADC sampling circuit and the constant current source circuit, and the result is displayed on the OLED screen.

[0067] This invention presents a crude oil water cut measurement method based on a GRU neural network model, which demonstrates excellent data processing performance. The GRU neural network model algorithm was used to predict the phase content of oil-water two-phase flow in a vertical riser pipe. Experimental data under various operating conditions were processed, and feature quantities were extracted in both the time and frequency domains. These were then analyzed using the GRU neural network model, yielding highly accurate water cut prediction results.

[0068] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A crude oil water content measurement device based on a GRU neural network model, characterized in that, The device includes a vertical five-electrode array sensor, a front-end signal processing and measurement circuit module, a microprocessor control module, and a temperature monitoring module. The vertical five-electrode array sensor is connected to the front-end signal processing and measurement circuit module, which is connected to the microprocessor control module. The microprocessor control module is connected to the temperature monitoring module, which is connected to the vertical five-electrode array sensor. The longitudinal five-electrode array sensor includes a cylinder (1), a pair of excitation electrodes (2), a pair of correlation velocity measurement electrodes (3), and an independent phase content measurement electrode (4) sleeved on the cylinder (1); the excitation electrodes (2) are located at both ends of the cylinder (1), the correlation velocity measurement electrodes (3) are located on the cylinder (1) inside the excitation electrodes (2), and the phase content measurement electrode (4) is located in the middle of the cylinder (1); the excitation electrodes (2) are connected to the front-end signal processing measurement circuit module; the correlation velocity measurement electrodes (3) are connected to the front-end signal processing measurement circuit module; and the phase content measurement electrode (4) is connected to both the front-end signal processing measurement circuit module and the temperature monitoring module. The front-end signal processing and measurement circuit module includes a signal generation circuit, a differential amplifier circuit, a limiting circuit, and an ADC sampling circuit; the signal generation circuit is connected to the excitation electrode (2) via the differential amplifier circuit; the differential amplifier circuit is connected to the limiting circuit, the limiting circuit is connected to the ADC sampling circuit, and the ADC sampling circuit is connected to the microprocessor control module. The microprocessor control module is equipped with a GRU neural network model, which is used to process the array of voltage elements acquired and output by the ADC sampling circuit to complete the measurement of phase content.

2. The crude oil water content measurement device based on the GRU neural network model according to claim 1, characterized in that, The excitation electrode (2), the related flow rate measuring electrode (3), and the phase content measuring electrode (4) are all made of stainless steel rings.

3. The crude oil water content measurement device based on the GRU neural network model according to claim 1, characterized in that, The microprocessor control module is a single-chip microcomputer. The microprocessor control module is connected to the ADC sampling circuit, and the other end of the ADC sampling circuit is connected to the limiting circuit. The microprocessor control module uses the cross-correlation velocity algorithm based on Fourier transform from the DSP library and the water content algorithm based on Fourier transform from the DSP library to measure the water content of crude oil.

4. The crude oil water content measurement device based on the GRU neural network model according to claim 1, characterized in that, The temperature monitoring module includes an analog output temperature sensor, a constant current source circuit, and an OLED screen; the analog output temperature sensor is connected to the constant current source circuit, and the OLED screen is connected to the microprocessor control module.

5. A method for measuring crude oil water content using a GRU neural network model-based measuring device according to any one of claims 1-4, characterized in that, Includes the following steps: S1. Apply an AC current with an amplitude of 5V and an output frequency of 18KHz to the excitation electrode of the longitudinal five-electrode array sensor as the excitation power supply for the longitudinal five-electrode array sensor. S2. When the fluid flows through the longitudinal five-electrode array sensor, the electric field generated by a pair of excitation electrodes produces a weak signal on the phase content measurement electrode. After being processed by the front-end signal processing measurement circuit module, the signal is sampled by the ADC circuit to obtain an array consisting of 4096 voltage elements. The array is then subjected to a fast Fourier transform based on the DSP library with 4096 points using the GRU neural network model. The crude oil water content information is concentrated in a certain fixed frequency band. The phase content measurement can be completed by calibrating based on the water content data. S3. A microprocessor control module is used to calculate the cross-correlation flow rate algorithm based on FFT acceleration and the moisture content algorithm based on FFT in the DSP library, and the results are output to the OLED screen of the temperature monitoring module. In step S2: the GRU neural network model is specifically as follows: S21. Training data: In the GRU neural network model, 4096 voltage elements collected by the ADC sampling circuit are used as the training set for training; the convolutional layer is set with a large convolutional kernel of 125×1. The GRU neural network model uses overlapping pooling with a pooling kernel length greater than the stride. The pooling kernel sizes are 5×1 and 3×1, and the stride of the two pooling layers is set to 2 to obtain a preliminary model. S22, Model Training: The essence of measuring moisture content data using a GRU neural network model is to extract key features from the corresponding samples for classification, and to use a logarithmic loss function to measure the difference between the predicted value and the true input value; the final loss function L of the GRU neural network model is defined, and the gradient of the hidden layer is: （1） The gradients corresponding to different parameter matrices can be calculated: （2） （3） （4） S23. Model Performance Evaluation: The GRU neural network model performs adaptive crude oil water content measurement on the collected phase content signals. First, the divided dataset is input into the GRU neural network model for training. The parameters are adjusted through the adaptive moment estimation algorithm to make the GRU neural network model converge. After obtaining the optimal training results, the model is saved. Then, crude oil water content is measured on the test set, and the diagnostic results are analyzed to verify the effectiveness of the GRU neural network model.

6. The method for measuring crude oil water content using a GRU neural network model according to claim 5, characterized in that, In step S3: The output result value is displayed on the OLED screen of the temperature monitoring module. Specifically, the temperature value detected by the analog output temperature sensor is sent to the microprocessor control module for processing through the ADC sampling circuit and the constant current source circuit, and the result is displayed on the OLED screen.

Images