Intelligent characterization method for shale oil imbibition front rate

Abstract

This invention discloses an intelligent characterization method for the velocity of the shale oil seepage front. The intelligent characterization method includes: constructing a terahertz imaging system; performing transmission scanning using the terahertz imaging system to obtain acquired data; preprocessing the acquired data to obtain a preprocessed terahertz image; obtaining a characterization result of the seepage front movement velocity based on the preprocessed terahertz image combined with a UNet-LSTM model; training the combined UNet-LSTM model; and inputting the preprocessed terahertz image into the trained combined UNet-LSTM model to obtain the final characterization result of the seepage front movement velocity. The use of multi-band terahertz imaging technology improves the resolution and penetration of the imaging; the combination of artificial intelligence algorithms enables automated data processing and analysis; and the optimization of imaging parameters through machine learning improves the accuracy and reliability of the characterization.

Description

Technical Field

[0001] This invention relates to the field of shale oil extraction technology, and in particular to an intelligent characterization method for the shale oil seepage front rate. Background Technology

[0002] Accurate characterization of the fracturing front rate is crucial for optimizing shale oil development strategies. This parameter directly affects the flow behavior of hydraulic fracturing fluids, fracture propagation efficiency, and ultimately, oil and gas recovery. However, due to the heterogeneity and complexity of shale reservoirs, obtaining this key parameter has become a pressing technical challenge.

[0003] Traditional imaging techniques, such as X-ray imaging (CT scan) and magnetic resonance imaging (MRI), have been widely used in shale oil extraction. These techniques can reveal the structural characteristics and fluid distribution of subsurface rocks to some extent, but they exhibit inherent limitations when dealing with materials like shale oil reservoirs, which have nanoscale porous structures. First, these imaging methods struggle to achieve the fine detail required to depict the minute pore structures; second, their penetration capabilities are also limited, especially when applied to deep or hard rocks, where image quality is often unsatisfactory. Therefore, while these techniques can provide macroscopic information, they offer limited assistance in gaining a deeper understanding of the microscopic processes within shale oil reservoirs.

[0004] Besides the inherent limitations of imaging technology itself, the reliance on manual operation in data processing and analysis is a significant problem. In practical applications, the entire process, from raw data acquisition to final result interpretation, requires substantial human intervention. This high dependence on manpower is not only inefficient but also increases the likelihood of errors. Especially when dealing with massive amounts of data, the difficulty and workload of manual analysis increase exponentially, further exacerbating the problem. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a smart characterization method for shale oil adsorption front rate that overcomes or at least partially solves the above problems.

[0006] According to one aspect of the present invention, a smart characterization method for shale oil adsorption front rates is provided, the smart characterization method comprising:

[0007] Constructing a terahertz imaging system;

[0008] Transmission scanning is performed using the terahertz imaging system to obtain the acquired data;

[0009] The collected data is preprocessed to obtain a preprocessed terahertz image;

[0010] Based on the preprocessed terahertz image combined with the UNet-LSTM model, the characterization results of the adsorption front movement velocity are obtained;

[0011] Training a joint UNet-LSTM model;

[0012] The preprocessed terahertz image is input into the trained joint UNet-LSTM model to obtain the final characterization result of the infiltration front movement velocity.

[0013] Optionally, constructing the terahertz imaging system specifically includes:

[0014] Select a pulse terahertz source and set the operating frequency range to [0.1, 3] THz;

[0015] Install a photoconductive antenna or an optical parametric oscillator (OPO) as a terahertz source;

[0016] Set up a terahertz imaging system, including a transmitter, a sample stage, and a receiver;

[0017] The sample stage is designed to be placed horizontally to ensure that the sample thickness is <1mm, and the distance between the sample stage and the terahertz wave source is adjusted to avoid imaging distortion.

[0018] Optionally, the step of performing a transmission scan using the terahertz imaging system to obtain the acquired data specifically includes:

[0019] Prepare shale oil samples;

[0020] Prepare the light source, set the imaging frequency range Frange = [0.1, 3] THz, select parallel light or focused light illumination mode, determine the exposure time Eexposure, keep the ambient temperature constant at Tenv, and determine the optimal exposure time, which is usually in the range of 0.1 to 500 ms to obtain the best image.

[0021] Multi-band imaging of shale oil samples was performed within a selected frequency range;

[0022] Point source preparation;

[0023] Teflon convex lens parallel light source;

[0024] Cylindrical lens line light source;

[0025] Scanning imaging;

[0026] The transmitted terahertz light is converted into a line source detection signal. The size of the detector tube matrix M×N and the sensitivity of the detector tube are set to ensure that the detection signal is transmitted to the detection screen composed of high-sensitivity detector tubes.

[0027] During the imaging process, ensure that the imaging area Area covers all areas to be detected, and control the imaging time Ttime to balance imaging quality and speed.

[0028] Optionally, the preparation of shale oil samples specifically includes:

[0029] Select representative shale oil samples to ensure that the porosity and saturation of the samples are known;

[0030] Control the water content of the sample to keep it consistent, and keep the sample thickness less than 1 mm to ensure that the terahertz wave can penetrate the sample without excessive attenuation.

[0031] Samples should be prepared at room temperature.

[0032] Optionally, the Teflon convex lens parallel light source is selected with a Teflon convex lens that is more than 10 times larger than the sample size, and the lens diameter and focal length are set to ensure that the terahertz wave can be converted into a parallel light source.

[0033] The cylindrical lens line light source uses a cylindrical lens that is more than 10 times larger than the sample. The length, radius and focal length of the cylindrical lens are set to ensure that a line light source that is more than 10 times larger than the sample size is formed at the focal point of the cylindrical lens.

[0034] The cylindrical lens should be positioned to ensure that the terahertz wave passes through the center of the cylindrical lens, forming a line light source that illuminates the sample.

[0035] Optionally, the scanning imaging specifically includes:

[0036] A high-speed deflecting mirror is added between the cylindrical lens and the focal point. The mirror area and deflection angle are set to ensure that the deflecting mirror can be adjusted to a certain angle to quickly scan the entire sample.

[0037] Optionally, the step of preprocessing the acquired data to obtain a preprocessed terahertz image specifically includes: normalizing the acquired data, removing noise, performing geometric correction, and enhancing image contrast through histogram equalization.

[0038] Optionally, obtaining the characterization result of the adsorption front movement velocity based on the preprocessed terahertz image and the UNet-LSTM model specifically includes:

[0039] The preprocessed terahertz image is input into the UNet model, and image features are extracted and segmentation masks are generated through convolutional layers and max pooling layers.

[0040] The segmented mask sequence is input into the LSTM model, and the long-term dependencies in the time series are captured by the LSTM unit to output the velocity estimate of the infiltration front.

[0041] The injection rate and volume of water-soluble modulating chemical reagents are controlled by a micro-injection system to ensure that the transmission spectrum of terahertz waves to the liquid is different, and the flow and distribution of the internal liquid are analyzed based on the transmission spectrum.

[0042] The flow of the internal liquid is analyzed based on the changes in the transmission spectrum. The changes in transmitted light intensity Δ, transmission wavelength λ, and absorption coefficient η are set to determine the distribution of the liquid in the sample.

[0043] The new terahertz image is input into the joint UNet-LSTM model. The UNet part extracts image features and generates segmentation masks, while the LSTM part predicts the moving velocity of the inhalation front based on the mask sequence, and outputs the final characterization result of the inhalation front moving velocity.

[0044] Optionally, training the joint UNet-LSTM model specifically includes:

[0045] Adjust hyperparameters based on model feedback, improve network architecture, select loss function, and enhance model generalization ability;

[0046] Record the effect of each modification during the optimization process;

[0047] The dataset is divided into training, validation, and test sets. The data split ratio is set, the evaluation metric is selected, the decision threshold is set, and cross-validation is performed to reduce bias.

[0048] Use multiple test sets to test the model, ensure the consistency of the test protocol, record test conditions to ensure test repeatability, and ensure that the test set data is diverse enough to cover a variety of situations.

[0049] Optionally, the step of inputting the preprocessed terahertz image into the trained joint UNet-LSTM model to obtain the final imbibition front movement velocity representation result specifically includes:

[0050] The terahertz image is input into the trained joint UNet-LSTM model;

[0051] UNet partially extracts image features and generates segmentation masks;

[0052] The LSTM part predicts the velocity of the adsorption front based on the mask sequence and outputs the final characterization result of the adsorption front velocity.

[0053] This invention provides an intelligent characterization method for the velocity of the shale oil seepage front. The intelligent characterization method includes: constructing a terahertz imaging system; performing transmission scanning using the terahertz imaging system to obtain acquired data; preprocessing the acquired data to obtain a preprocessed terahertz image; obtaining a characterization result of the seepage front movement velocity based on the preprocessed terahertz image combined with a UNet-LSTM model; training the combined UNet-LSTM model; and inputting the preprocessed terahertz image into the trained combined UNet-LSTM model to obtain the final characterization result of the seepage front movement velocity. The use of multi-band terahertz imaging technology improves the resolution and penetration of the imaging; the combination of artificial intelligence algorithms enables automated data processing and analysis; and the optimization of imaging parameters through machine learning improves the accuracy and reliability of the characterization.

[0054] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A flowchart illustrating an intelligent characterization method for shale oil adsorption front rate provided in an embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of the composition of a terahertz imaging device provided in an embodiment of the present invention. Detailed Implementation

[0058] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0059] The terms "comprising" and "having," and any variations thereof, in the specification, embodiments, claims, and drawings of this invention are intended to cover non-exclusive inclusion, such as including a series of steps or units.

[0060] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0061] Example 1

[0062] like Figure 1 As shown, a smart characterization method for shale oil adsorption front rate includes:

[0063] A terahertz imaging system is constructed; transmission scanning is performed using the terahertz imaging system to obtain acquired data; the acquired data is preprocessed to obtain a preprocessed terahertz image; based on the preprocessed terahertz image and a UNet-LSTM model, the velocity characterization result of the adsorption front is obtained; the joint UNet-LSTM model is trained; the preprocessed terahertz image is input into the trained joint UNet-LSTM model to obtain the final velocity characterization result of the adsorption front.

[0064] 1. Construct a terahertz imaging system, the specific structure of which is shown in the diagram below. Figure 2 As shown. First, select a pulsed terahertz source and set the operating frequency range to [0.1, 3] THz. Install a photoconductive antenna or optical parametric oscillator (OPO) as the terahertz source, ensuring it can generate a stable pulse signal. Design and build a terahertz imaging system, including a transmitter, sample stage, and receiver. The sample stage is designed to be horizontally placed, ensuring the sample thickness is less than 1 mm, and adjust the distance between the sample stage and the terahertz source to ensure it is appropriate, avoiding excessive proximity or distance that could cause imaging distortion.

[0065] 2. Prepare shale oil samples. Select representative shale oil samples, ensuring that the porosity and saturation are known. Control the water content of the samples to maintain consistency, and the sample thickness should be less than 1 mm to ensure that terahertz waves can penetrate the sample without excessive attenuation. Samples should be prepared at room temperature to avoid the influence of temperature fluctuations on sample characteristics.

[0066] 3. Prepare the light source. Set the imaging frequency range to [0.1, 3] THz, select parallel light or focused light illumination mode, determine the exposure time, and keep the ambient temperature constant at Tenv to ensure that the ambient temperature will not affect the sample characteristics and equipment performance during imaging. Determine the optimal exposure time through experiments to obtain the best image quality.

[0067] 4. Multi-band selection. Perform multi-band imaging of shale oil samples within the selected frequency range, ensuring the use of Nbands = 5 different frequency bands [using 1-30 bands, preferably 2-15 bands], and adjust the spatial resolution [10-200µm, preferably 20-80µm] to ensure image stitching accuracy. Adjust the viewpoint angle to obtain sample information from different perspectives. Ensure appropriate frequency intervals for each imaging session to avoid interference between frequency bands.

[0068] 5. Point Source Preparation. Waveguide technology is employed to convert the terahertz wave into a point source. The waveguide length, width, and height are designed to ensure lossless transmission of the terahertz wave to the sample location. Low-loss materials are selected for the waveguide to maximize signal strength. The waveguide outlet is designed according to the sample size to ensure precise illumination of the sample.

[0069] 6. Teflon Convex Lens Parallel Light Source. Select a Teflon convex lens that is at least 10 times larger than the sample size, and set the lens diameter and focal length to ensure that the terahertz waves can be converted into a parallel light source. Teflon material has excellent transmission properties and low absorption, which can effectively reduce energy loss. The lens is installed at the waveguide outlet to ensure that the beam can uniformly illuminate the sample after passing through the lens.

[0070] 7. Cylindrical Lens Line Light Source. Select a cylindrical lens at least 10 times larger than the sample. Set the length, radius, and focal length of the cylindrical lens to ensure that a line light source at least 10 times larger than the sample size is formed at the focal point of the cylindrical lens. The installation position of the cylindrical lens should ensure that the terahertz wave can pass through the center of the cylindrical lens, forming a line light source that illuminates the sample.

[0071] 8. Rapid Scanning. A high-speed deflecting mirror is added between the cylindrical lens and its focal point. The mirror area [Amirror, 2-400 cm², preferably 5-160 cm²] and deflection angle [Ttilt, 1-30°, preferably 2-8°] are set to ensure that the deflecting mirror only needs a small angle adjustment to achieve rapid scanning of the entire sample. The mirror's drive system should have high precision and high response speed to ensure stability and speed during the scanning process.

[0072] 9. After transmitting the terahertz light, design a detection system to collect information. Set the detector diameter [2-30 cm, preferably 3-15 cm] and the number of detectors [20-700, preferably 40-300] to ensure the detector array area can cover the sample. The detectors should be made of highly sensitive materials to ensure the detection of even weak terahertz signals. The detection system should be mounted behind the cylindrical lens to ensure accurate signal collection.

[0073] 10. Convert the transmitted terahertz light into a line source detection signal. Set the detector array size M×N [M ranges from 1-60, preferably 5-25; N ranges from 1-200, preferably 5-40] and the detector sensitivity to ensure the detection signal is transmitted to the detection screen composed of high-sensitivity detectors. The spacing between the detectors should be appropriate to ensure detection efficiency without mutual interference. The detection screen should be installed at the end of the transmission path to ensure complete signal capture.

[0074] 11. During imaging, ensure the imaging area covers the entire area to be detected, and control the imaging time [1-600 seconds, preferably 2-120 seconds] to balance imaging quality and speed. Optimize image quality by adjusting imaging parameters and eliminate background interference. During data acquisition, ensure system stability and avoid the impact of external vibrations on imaging quality.

[0075] 12. Normalize the collected data, remove noise, perform geometric correction, and enhance image contrast through histogram equalization. The effect of each step in the preprocessing process should be carefully checked to ensure that the data quality meets the requirements of subsequent analysis.

[0076] 13. Input the preprocessed terahertz image into the UNet model, extract image features through convolutional and max-pooling layers, and generate segmentation masks. Then, input the segmentation mask sequence into the LSTM model, capture long-term dependencies in the time series using LSTM units, and finally output the velocity estimate of the seepage front. Intermediate results should be saved periodically during model training for subsequent debugging and optimization.

[0077] 14. Water and water-soluble modulating chemicals, such as water-soluble phenolic resin crosslinking agents, are precisely injected using a micro-injection system at controlled rates (0.1-20 μL) and volumes (depending on core porosity). This ensures that the terahertz wave transmission spectra differ for these liquids, allowing analysis of internal liquid flow and distribution based on transmission spectroscopy. Before injection, the sample should be kept stationary to avoid disturbance during injection that could affect imaging quality.

[0078] 15. Analyze the internal liquid flow based on changes in the transmission spectrum. Set the transmitted light intensity change Δ, transmission wavelength λ, and absorption coefficient η to determine the liquid distribution in the sample. Professional software tools should be used for transmission spectroscopy analysis to ensure the accuracy and reliability of the results.

[0079] 16. Input the new terahertz image into the joint UNet-LSTM model. The UNet part extracts image features and generates segmentation masks, while the LSTM part predicts the moving velocity of the infiltration front based on the mask sequence. Finally, the model outputs the final representation of the infiltration front moving velocity. Before inputting the image into the model, ensure that the image format is consistent to avoid errors caused by inconsistent formats.

[0080] 17. Adjust hyperparameters based on model feedback, improve network architecture, select an appropriate loss function (Lloss), and enhance model generalization ability to prevent overfitting. Record the effect of each modification during optimization to ensure each improvement delivers performance gains.

[0081] 18. Divide the dataset into training, validation, and test sets, set the data split ratio, select evaluation metrics, set decision thresholds, and perform cross-validation to reduce bias. During validation, variables should be strictly controlled to ensure the comparability of validation results for each iteration.

[0082] 19. Use a large-scale test set for model testing to ensure consistency of the testing protocol, record test conditions to guarantee test repeatability, and ensure that the test set data is diverse enough to cover multiple scenarios. During testing, ensure consistency in data processing to avoid errors caused by differences in processing.

[0083] 20. Using a pre-trained joint UNet-LSTM model, a new terahertz image is input into the model. The UNet part extracts image features and generates segmentation masks, while the LSTM part predicts the moving velocity of the infiltration front based on the mask sequence. Finally, the final infiltration front moving velocity representation result is output. A final check should be performed before outputting the results to ensure the accuracy and completeness of the output.

[0084] This invention employs multi-band terahertz imaging technology, which improves the resolution and penetration of the imaging; combined with artificial intelligence algorithms, it realizes automated data processing and analysis; and by optimizing imaging parameters through machine learning, it improves the accuracy and reliability of characterization.

[0085] The present invention provides an intelligent characterization method for shale oil seepage front rate, which can more accurately characterize shale oil seepage front rate, improve mining efficiency, reduce costs, and reduce the complexity of manual operation.

[0086] Example 2

[0087] The embodiments of the present invention include: imaging shale oil samples using a multi-band terahertz imaging system to obtain imaging data in different frequency bands; inputting the imaging data into an artificial intelligence algorithm for data processing and analysis; optimizing imaging parameters using machine learning technology to improve the accuracy of characterization; and outputting intelligent characterization results of the infiltration front rate.

[0088] The specific steps include:

[0089] Select a pulsed terahertz source and set the operating frequency to the range of 0.1-3THz, with 1.5THz being preferred. Install a photoconductive antenna as the terahertz source to ensure that it can generate a stable pulse signal. Design and build a terahertz imaging system, including a transmitter, a sample stage, and a receiver. The sample stage is designed to be placed horizontally, ensuring that the sample thickness is less than 0.8mm. Adjust the sample stage to ensure that the distance between it and the terahertz source is appropriate at 20cm to avoid imaging distortion caused by being too close or too far.

[0090] A representative shale oil sample was selected, with a known porosity of 0.2 and a saturation of 0.7. The water content of the sample was kept constant at 0.15, and the sample thickness was less than 0.8 mm to ensure that terahertz waves could penetrate the sample without excessive attenuation. The sample should be prepared at room temperature to avoid the influence of temperature fluctuations on the sample characteristics.

[0091] The imaging frequency was set to 1.5 THz, the parallel light illumination mode was selected, the exposure time was set to 50 ms, and the ambient temperature was kept constant at Tenv = 25℃ to ensure that the ambient temperature would not affect the sample characteristics and equipment performance during the imaging process. Multi-band imaging of shale oil samples was performed within the selected frequency range, ensuring that Nbands = 5 different frequency bands were used, the spatial resolution was adjusted to 50 μm, the scan step size was set to 1 mm to ensure the accuracy of image stitching, and the viewpoint angle was adjusted to 45° to obtain sample information from different perspectives.

[0092] To convert terahertz waves into a point source, waveguide technology was employed. The waveguide was designed with a length of 50cm, a width of 2cm, and a height of 2cm to ensure that the terahertz waves could be transmitted to the sample location without loss. Low-loss materials were selected for the waveguide to maximize signal strength. The waveguide outlet was designed according to the sample size to ensure that it could accurately irradiate the sample.

[0093] Select a Teflon convex lens that is more than 10 times larger than the sample size, and set the lens diameter to 50cm and the focal length to 30cm to ensure that the terahertz wave can be converted into a parallel light source. Teflon material has excellent transmission performance and low absorption rate, which can effectively reduce energy loss. The lens is installed at the waveguide outlet to ensure that the beam can be evenly irradiated onto the sample after passing through the lens.

[0094] Select a cylindrical lens that is more than 10 times larger than the sample. Set the length of the cylindrical lens to 50cm, the radius to 25cm, and the focal length to 30cm. Ensure that a line light source that is more than 10 times larger than the sample size is formed at the focal point of the cylindrical lens. The installation position of the cylindrical lens should ensure that the terahertz wave can pass through the center of the cylindrical lens and illuminate the sample after forming a line light source.

[0095] A high-speed deflecting mirror is added between the cylindrical lens and its focal point, with the mirror area set to Amirror = 100 cm². 2 With a deflection angle Ttilt = 5°, the deflecting mirror only needs to be adjusted by a small angle to achieve rapid scanning of the entire sample. The driving system of the mirror should have high precision and high response speed to ensure stability and speed during the scanning process.

[0096] After transmitting terahertz light, a detection system is designed to collect information. The detector diameter is set to 5 cm and the number of detectors is set to 100, ensuring that the detector array area Aarray = 50 cm². 2 To ensure the detector can cover the sample, a high-sensitivity material should be selected to ensure that weak terahertz signals can be detected. The detection system should be installed behind the cylindrical lens to ensure that the signal can be collected accurately.

[0097] The transmitted terahertz light is converted into a line source detection signal. The size of the detector tube matrix is ​​set to M×N=10×10 and the detector tube sensitivity is set to 0.5mV / mW. This ensures that the detection signal is transmitted to the detection screen composed of high-sensitivity detector tubes. The spacing between the detector tubes should be appropriate to ensure detection efficiency without mutual interference. The detection screen should be installed at the end of the transmission path to ensure that the signal can be completely captured.

[0098] During the imaging process, ensure that the imaging area Aarea = 50cm. 2 Cover all regions of interest, control the imaging time Ttime = 10s, optimize image quality by adjusting imaging parameters and eliminate background interference. During data acquisition, pay attention to maintaining system stability and avoid the impact of external vibration on image quality.

[0099] The collected data is normalized, noise is removed, and geometric correction is performed. Histogram equalization is used to enhance image contrast. The effect of each step in the preprocessing process should be carefully checked to ensure that the data quality meets the requirements of subsequent analysis. The preprocessed terahertz image is input into the UNet model. Image features are extracted through convolutional layers and max pooling layers to generate segmentation masks. The segmentation mask sequence is then input into the LSTM model. The LSTM unit captures long-term dependencies in the time series and finally outputs the velocity estimate of the osmosis front. Intermediate results should be saved regularly during model training to facilitate subsequent debugging and optimization. Water and water-soluble phenolic resin crosslinking agents and other water-soluble modulating chemicals are injected at a precise controlled rate and volume through a micro-injection system to ensure that the terahertz wave transmits different spectra to these liquids. The internal liquid flow and distribution can be analyzed based on the transmission spectrum. Before injecting the liquid, the sample should be kept still to avoid disturbance during the liquid injection process from affecting the imaging quality.

[0100] The flow of the internal liquid is analyzed based on the changes in the transmission spectrum. The transmission light intensity change Δ = 0.2, the transmission wavelength λ = 1.5 μm, and the absorption coefficient η = 0.05 are set to determine the distribution of the liquid in the sample. Professional software tools should be used for transmission spectrum analysis to ensure the accuracy and reliability of the analysis results. The new terahertz image is input into the joint UNet-LSTM model. The UNet part extracts image features and generates a segmentation mask, while the LSTM part predicts the moving velocity of the adsorption front based on the mask sequence. Finally, the final adsorption front moving velocity characterization result is output. Before inputting the model, the image format should be consistent to avoid errors caused by inconsistent formats.

[0101] Adjust hyperparameters based on model feedback, improve network architecture, select an appropriate loss function Lloss, and enhance model generalization ability to prevent overfitting. Record the effect of each modification during optimization to ensure that each improvement brings performance gains. Divide the dataset into training, validation, and test sets, setting the data split ratio Ssplit = 70%. Select evaluation metrics, set the decision threshold Tthreshold = 0.5, and perform cross-validation Ccross = 5 to reduce bias. Strictly control variables during validation to ensure the comparability of validation results.

[0102] The model was tested using a large-scale test set Ttest=1000 to ensure consistency of the testing protocol. Test conditions were recorded to guarantee repeatability, and the test set data was designed to cover a variety of scenarios. Consistency in data processing was maintained during testing to avoid errors caused by processing differences. A new terahertz image was input into the trained joint UNet-LSTM model. The UNet part extracted image features and generated segmentation masks, while the LSTM part predicted the velocity of the water absorption front based on the mask sequence. The final output was a representation of the absorption front velocity. A final check was performed before output to ensure the accuracy and completeness of the results. Based on this patent, the accuracy of water absorption front analysis was improved by more than 12% compared to the staining method.

[0103] Beneficial effects:

[0104] This invention discloses an intelligent characterization method for shale oil seepage front velocities based on multi-band terahertz imaging technology. This method addresses the problems of insufficient accuracy, low efficiency, and complex data processing in existing shale oil extraction technologies for characterizing seepage front velocities. The core of this invention lies in utilizing multi-band terahertz imaging technology combined with advanced artificial intelligence algorithms to achieve high-precision and high-efficiency intelligent characterization of shale oil seepage front velocities.

[0105] This invention first uses a multi-band terahertz imaging system to image shale oil samples, acquiring imaging data of the samples at different frequency bands. This data is then input into a data processing system integrating artificial intelligence algorithms, which can automatically process and analyze the data to extract key information about the adsorption front. To further improve the accuracy and reliability of the characterization, this invention also employs machine learning techniques to optimize imaging parameters and analysis algorithms.

[0106] Compared with existing technologies, this invention has significant advantages: First, multi-band terahertz imaging technology provides higher imaging resolution and penetration, making the characterization of shale oil seepage fronts more accurate; second, the integrated artificial intelligence algorithm greatly reduces the need for manual operation and improves the efficiency of data processing; finally, the imaging parameters and analysis algorithms optimized by machine learning further improve the accuracy and reliability of the characterization results.

[0107] The intelligent characterization method of this invention is not only applicable to various shale oil extraction scenarios, especially shale oil extraction in complex environments such as deep formations, high pressure, and high temperature, but also of great significance for improving extraction efficiency, reducing costs, and reducing the complexity of manual operations.

[0108] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart characterization method for shale oil adsorption front rate, characterized in that, The intelligent representation method includes: Constructing a terahertz imaging system; Transmission scanning is performed using the terahertz imaging system to obtain the acquired data; The collected data is preprocessed to obtain a preprocessed terahertz image; Based on the preprocessed terahertz image combined with the UNet-LSTM model, the characterization results of the adsorption front movement velocity are obtained; Training a joint UNet-LSTM model; The preprocessed terahertz image is input into the trained joint UNet-LSTM model to obtain the final characterization result of the infiltration front movement velocity.

2. The intelligent characterization method for shale oil adsorption front rate according to claim 1, characterized in that, The construction of the terahertz imaging system specifically includes: Select a pulse terahertz source and set the operating frequency range to [0.1, 3] THz; Install a photoconductive antenna or an optical parametric oscillator (OPO) as a terahertz source; Set up a terahertz imaging system, including a transmitter, a sample stage, and a receiver; The sample stage is designed to be placed horizontally to ensure that the sample thickness is <1mm, and the distance between the sample stage and the terahertz wave source is adjusted to avoid imaging distortion.

3. The intelligent characterization method for shale oil adsorption front rate according to claim 1, characterized in that, The specific steps of obtaining the acquired data by performing a transmission scan using the terahertz imaging system include: Prepare shale oil samples; Prepare the light source, set the imaging frequency range Frange = [0.1, 3] THz, select parallel light or focused light illumination mode, determine the exposure time, keep the ambient temperature constant at Tenv, determine the optimal exposure time, and obtain the best image; Multi-band imaging of shale oil samples was performed within a selected frequency range; Point source preparation; Teflon convex lens parallel light source; Cylindrical lens line light source; Scanning imaging; The transmitted terahertz light is converted into a line source detection signal. The size of the detector tube matrix M×N and the sensitivity of the detector tube are set to ensure that the detection signal is transmitted to the detection screen composed of high-sensitivity detector tubes. During the imaging process, ensure that the imaging area Area covers all areas to be detected, and control the imaging time Ttime to balance imaging quality and speed.

4. The intelligent characterization method for shale oil adsorption front rate according to claim 3, characterized in that, The preparation of shale oil samples specifically includes: Select representative shale oil samples to ensure that the porosity and saturation of the samples are known; Control the water content of the sample to keep it consistent, and keep the sample thickness less than 1 mm to ensure that the terahertz wave can penetrate the sample without excessive attenuation. Samples should be prepared at room temperature.

5. The intelligent characterization method for shale oil adsorption front rate according to claim 3, characterized in that, The Teflon convex lens parallel light source selects a Teflon convex lens that is more than 10 times larger than the sample size, sets the lens diameter and focal length, and ensures that the terahertz wave can be converted into a parallel light source. The cylindrical lens line light source uses a cylindrical lens that is more than 10 times larger than the sample. The length, radius and focal length of the cylindrical lens are set to ensure that a line light source that is more than 10 times larger than the sample size is formed at the focal point of the cylindrical lens. The cylindrical lens should be positioned to ensure that the terahertz wave passes through the center of the cylindrical lens, forming a line light source that illuminates the sample.

6. The intelligent characterization method for shale oil adsorption front rate according to claim 3, characterized in that, The scanning imaging specifically includes: A high-speed deflecting mirror is added between the cylindrical lens and the focal point. The mirror area and deflection angle are set to ensure that the deflecting mirror can be adjusted to a certain angle to quickly scan the entire sample.

7. The intelligent characterization method for shale oil adsorption front rate according to claim 1, characterized in that, The step of preprocessing the acquired data to obtain a preprocessed terahertz image specifically includes: The collected data is normalized, noise is removed, geometric correction is performed, and image contrast is enhanced through histogram equalization.

8. The intelligent characterization method for shale oil adsorption front rate according to claim 1, characterized in that, The specific steps for obtaining the characterization results of the adsorption front movement velocity based on the preprocessed terahertz image combined with the UNet-LSTM model include: The preprocessed terahertz image is input into the UNet model, and image features are extracted and segmentation masks are generated through convolutional layers and max pooling layers. The segmented mask sequence is input into the LSTM model, and the long-term dependencies in the time series are captured by the LSTM unit to output the velocity estimate of the infiltration front. The injection rate and volume of water-soluble modulating chemical reagents are controlled by a micro-injection system to ensure that the transmission spectrum of terahertz waves to the liquid is different, and the flow and distribution of the internal liquid are analyzed based on the transmission spectrum. The flow of the internal liquid is analyzed based on the changes in the transmission spectrum. The changes in transmitted light intensity Δ, transmission wavelength λ, and absorption coefficient η are set to determine the distribution of the liquid in the sample. The new terahertz image is input into the joint UNet-LSTM model. The UNet part extracts image features and generates segmentation masks, while the LSTM part predicts the moving velocity of the inhalation front based on the mask sequence, and outputs the final characterization result of the inhalation front moving velocity.

9. The intelligent characterization method for shale oil adsorption front rate according to claim 1, characterized in that, The training of the joint UNet-LSTM model specifically includes: Adjust hyperparameters based on model feedback, improve network architecture, select loss function, and enhance model generalization ability; Record the effect of each modification during the optimization process; The dataset is divided into training, validation, and test sets. The data split ratio is set, evaluation metrics are selected, decision thresholds are set, and cross-validation is performed to reduce bias. Use multiple test sets to test the model, ensure the consistency of the test protocol, record test conditions to ensure test repeatability, and ensure that the test set data is diverse enough to cover a variety of situations.

10. The intelligent characterization method for shale oil adsorption front rate according to claim 1, characterized in that, The step of inputting the preprocessed terahertz image into the trained joint UNet-LSTM model to obtain the final imbibition front movement velocity representation result specifically includes: The terahertz image is input into the trained joint UNet-LSTM model; UNet partially extracts image features and generates segmentation masks; The LSTM part predicts the velocity of the adsorption front based on the mask sequence and outputs the final characterization result of the adsorption front velocity.

Images