Nuclear magnetic resonance fast relaxation component signal enhancement system and method based on residual neural network

Abstract

The invention discloses a nuclear magnetic resonance fast relaxation component signal enhancement system and method based on a residual neural network. The system comprises low-field nuclear magnetic resonance equipment and a main control computer, the low-field nuclear magnetic resonance equipment comprises a signal acquisition system, and a ResNet module is embedded in the low-field nuclear magnetic resonance equipment; the low-field nuclear magnetic resonance equipment is used for exciting a tested sample to generate a fast relaxation signal through a CPMG pulse; collecting multi-dimensional echo data by using multiple channels, carrying out feature fusion on the multi-channel data through a ResNet module, and transmitting the multi-channel data to a main control computer; and the main control computer is used for acquiring real-time signals of the signal acquisition system, and dynamically adjusting acquisition parameters and processing strategies based on physical characteristics of the tested sample and a feedback control mechanism so as to maintain the stability of the signals. According to the method, the residual learning capability of ResNet is utilized, and multi-channel acquisition and dynamic parameter modulation are combined, so that the amplitude and the signal-to-noise ratio of the fast relaxation signal are improved, and an efficient technical scheme is provided for shale oil and gas reservoir characterization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of shale oil and gas characterization, and particularly relates to a nuclear magnetic resonance (NMR) fast relaxation component signal enhancement system and method based on a residual neural network. BACKGROUND

[0002] NMR technology can non-destructively detect fluid content and pore structure in a reservoir by measuring the relaxation characteristics of hydrogen nuclei, and has a wide application in the field of oil and gas exploration, especially in the characterization of complex reservoirs such as shale oil and gas. However, the transverse relaxation time (T2) of fast relaxation component signals (such as clay bound water and solid organic matter) is extremely short, usually less than 1 ms, and rapidly decays in the initial echo phase of the CPMG (Carr-Purcell-Meiboom-Gill) pulse sequence, resulting in extremely low signal intensity. For example, in a low-field NMR device (magnetic field strength < 0.1 T), the initial amplitude is often less than 10 μV, and because of the rapid decay, the sampling points are sparse, making it difficult to meet the demand for high-intensity signals for precise characterization of complex reservoirs. This signal weakening problem is particularly pronounced in unconventional reservoirs such as shale, as the micro-nano pore structure makes the proportion of fast relaxation components higher.

[0003] To enhance the signal of fast relaxation components, existing technologies mainly focus on hardware and sequence design. Increasing the magnetic field strength or the power of the radio frequency pulse can increase the initial amplitude of the signal, but high-field devices are expensive and not portable, and power increases are limited by instrument power consumption and sample tolerance, limiting their practical application. Another common method is to shorten the echo interval (TE), such as from 0.8 ms to 0.2 ms, to increase the front-end sampling points, but the signal front end is still prone to loss due to the dead time of the radio frequency coil (usually > 0.1 ms), and the insufficient response speed of the hardware may introduce distortion. However, existing NMR techniques for enhancing fast relaxation component signals mainly rely on hardware upgrades or simple sequence adjustments, and do not fully utilize data processing algorithms to optimize signals, resulting in limited signal enhancement and difficulty in meeting the needs of precise characterization of complex reservoirs.

[0004] Therefore, the present application proposes a NMR fast relaxation component signal enhancement method based on a residual neural network (ResNet), which utilizes the residual learning ability of ResNet to enhance the signal of fast relaxation components from both acquisition optimization and data processing, providing technical support for the precise characterization of complex reservoirs such as shale oil and gas. SUMMARY

[0005] In view of the above, the present application provides a residual neural network-based nuclear magnetic resonance fast-relaxation component signal enhancement system and method to solve the technical problem that the existing method for enhancing fast-relaxation component signals is subject to cost, power consumption and hardware limits, resulting in limited signal enhancement effect and difficulty in adapting to complex reservoir diversity in low-field environments.

[0006] To achieve the above technical purpose, the present application adopts the following technical solutions: In a first aspect, the present application provides a residual neural network-based nuclear magnetic resonance fast-relaxation component signal enhancement system, comprising a low-field nuclear magnetic resonance device and a host computer connected to each other; wherein the low-field nuclear magnetic resonance device comprises a signal acquisition system, the signal acquisition system is connected to the host computer through a high-speed data interface, and a pre-trained ResNet module is embedded therein; The low-field nuclear magnetic resonance device is configured to configure a CPMG pulse, excite a measured sample to generate a fast-relaxation signal through the CPMG pulse, acquire multi-dimensional echo data through multi-channel parallel acquisition, perform feature fusion on the multi-channel data through the ResNet module to improve the signal amplitude, and transmit the signal with the improved amplitude to the host computer; The host computer is configured in a real-time processing mode, used to acquire real-time signals of the signal acquisition system, dynamically adjust acquisition parameters and processing strategies based on physical characteristics of the measured sample and a feedback control mechanism, and maintain the stability of the signals.

[0007] Further, the ResNet module comprises a plurality of residual blocks, each residual block comprising at least two convolution layers, the convolution kernel size of each convolution layer being arranged along the signal time domain direction, and the number of channels of each convolution layer being doubled layer by layer, used to extract time domain features of the multi-channel signal and complete feature fusion.

[0008] Further, the signal acquisition system comprises a multi-phase channel and a multi-frequency offset channel; The phases of the multi-phase channel are configured to be orthogonal to each other, and the frequency offsets of the multi-frequency offset channel are configured to be symmetrically offset; the multi-phase channel and the multi-frequency offset channel are used to synchronously acquire fast-relaxation component signals to generate a multi-dimensional initial echo data set.

[0009] Further, the host computer comprises a dynamic modulation unit and a monitoring feedback unit; The dynamic modulation unit is used to dynamically adjust the radio frequency parameters of the CPMG pulse sequence based on the physical characteristics of the sample and the real-time signal feedback; The monitoring feedback unit is used to monitor the amplitude and decay characteristics of the enhanced signal in real time, and dynamically adjust the acquisition parameters and processing strategies through a feedback mechanism.

[0010] In another aspect, the present application also provides a method for enhancing the signal of a fast-relaxation component in nuclear magnetic resonance based on a residual neural network, which is applied to any of the systems for enhancing the signal of a fast-relaxation component in nuclear magnetic resonance based on a residual neural network described in the above technical solutions and comprises the following steps of: Calibrating initial CPMG pulse parameters of a low-field nuclear magnetic resonance device; Obtaining a core sample, optimizing a CPMG pulse sequence according to a pre-experiment result of the core sample, and determining optimal pulse power; Performing nuclear magnetic scanning on the core sample based on the optimized CPMG pulse sequence, and inputting signals of the fast-relaxation component into a pre-trained ResNet module through multi-channel parallel acquisition; Extracting and fusing time-frequency features of the multi-channel signals by using the ResNet module, superimposing effective signal components, and outputting the enhanced signals of the fast-relaxation component; Dynamically adjusting pulse power and echo interval parameters of the CPMG pulse sequence according to amplitudes and decay rates of the signals of the fast-relaxation component, and inputting the modulated signals into the ResNet module for secondary enhancement; Real-time monitoring of amplitudes and decay characteristics of the enhanced signals, dynamic adjustment of acquisition parameters and processing strategies in combination with a feedback control mechanism, and keeping the amplitudes of the enhanced signals stable.

[0011] Further, the initial CPMG pulse parameters include radio frequency pulse power, initial phase, and echo interval.

[0012] Further, the extraction and fusion of time-frequency features of the multi-channel signals by using the ResNet module comprises the following steps of: Extracting time-domain features of the multi-channel data, and separating time-frequency features of the fast-relaxation signals through short-time Fourier transform; Inputting the extracted feature vectors into the ResNet module, and jointly analyzing and optimizing the nuclear magnetic resonance signal data from different channels by using a residual learning path to enhance the detectability of the signals in a low signal-to-noise ratio environment.

[0013] Further, the dynamic modulation of the parameters of the CPMG pulse sequence comprises the following steps of: Adjusting a step value of the echo interval according to a real-time signal amplitude, and adjusting the power of the CPMG pulse sequence in combination with the porosity of the measured sample.

[0014] Further, the feedback control mechanism comprises the following steps of: Monitoring amplitudes and decay rates of the enhanced signals; When the amplitude is lower than a preset threshold, increasing the number of phase channels or expanding the frequency offset range to maintain the stability of the signals; and when the decay rate exceeds a preset speed, shortening the echo interval and synchronously updating the input of the ResNet.

[0015] Further, the CPME pulse sequence is optimized according to the pre-experiment results of the core sample, and the optimization of the CPME pulse sequence includes: Step-by-step increase of the 90° pulse power, determination of the amplitude variation of a preset number of echo points in accordance with the first preset standard; Setting the 180° pulse power, verifying that the signal decay rate of the 180° pulse meets the second preset standard; Real-time monitoring of the signal waveform by an oscilloscope to ensure that the decay rate is controlled within a preset range.

[0016] Compared with the prior art, the residual neural network-based nuclear magnetic resonance fast relaxation component signal enhancement system and method has the following advantages: 1. Enhancing the intensity and signal-to-noise ratio of fast relaxation signals: Through residual learning, the front-end echo amplitude of fast relaxation signals is amplified in real time, and effective signal components are extracted and superimposed to suppress noise interference; using multi-phase and multi-frequency channel data, combined with ResNet, to enhance the detectability of signals in low signal-to-noise ratio environments.

[0017] 2. High real-time performance and low delay processing capability: ResNet modules are embedded in the signal acquisition system through high-speed data interfaces, supporting real-time processing and adapting to the fast acquisition requirements of nuclear magnetic resonance signals.

[0018] 3. Dynamic adaptability optimization and stability: Through dynamic modulation of acquisition parameters and ResNet collaborative optimization, a self-adaptive fast relaxation signal enhancement mechanism is formed. It can dynamically adjust the CPMG pulse parameters according to the sample characteristics and real-time signal feedback; at the same time, by monitoring the amplitude and decay characteristics of the enhanced signal, the number of phase channels or the frequency offset range is automatically expanded to ensure signal stability.

[0019] In summary, the system of the present application realizes the enhancement of fast relaxation signals, noise suppression and signal stability optimization through multi-channel feature fusion, real-time processing capability and dynamic adaptability optimization, enhances the fast relaxation component signal from the aspects of acquisition optimization and data processing, and provides technical support for accurate characterization of complex reservoirs such as shale oil and gas. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The structure diagram of the residual neural network-based nuclear magnetic resonance fast relaxation component signal enhancement system provided by the present application is provided; Figure 2 The flowchart of the residual neural network-based nuclear magnetic resonance fast relaxation component signal enhancement method provided by the present application is provided. DETAILED DESCRIPTION

[0021] The preferred embodiments of the present application will be described in detail below with reference to the drawings, wherein the drawings constitute a part of this application, and are used to explain the principles of the embodiments of the present application, but are not used to limit the scope of the present application.

[0022] Embodiment 1 Please refer to Figure 1 The embodiment provides a residual neural network-based signal enhancement system for fast-relaxation components of nuclear magnetic resonance 100, comprising a low-field nuclear magnetic resonance device 101 and a host computer 102 connected to each other; wherein the low-field nuclear magnetic resonance device 101 comprises a signal acquisition system 103, the signal acquisition system 103 is connected with the host computer 102 through a high-speed data interface, and a pre-trained ResNet module 104 is embedded in the signal acquisition system 103. The low-field nuclear magnetic resonance device 101 is used to configure a CPMG pulse, excite a sample to be measured to generate a fast-relaxation signal through the CPMG pulse, acquire multi-dimensional echo data through multi-channel parallel acquisition, perform feature fusion on the multi-channel data through the ResNet module 104, and improve the signal amplitude; and the signal with the improved amplitude is transmitted to the host computer 102. The host computer 102 is configured in a real-time processing mode, is used to acquire real-time signals of the signal acquisition system 103, dynamically adjusts acquisition parameters and processing strategies based on sample physical characteristics and a feedback control mechanism, and maintains the stability of the signals.

[0023] The residual neural network-based signal enhancement system for fast-relaxation components of nuclear magnetic resonance provided in the embodiment can greatly improve the quality of the signals by embedding a pre-trained ResNet module to perform feature fusion on echo data acquired through multi-channel acquisition, thereby effectively improving the amplitude of the fast-relaxation signals, and enabling the fast-decaying signals to be more clearly captured and analyzed. The host computer processes the signals in real time and dynamically adjusts the acquisition parameters and the processing strategies, and based on the sample physical characteristics and the feedback control mechanism, the system can automatically adjust various settings in the acquisition process, thereby maintaining the stability of the signals under different experimental conditions. Since the system can automatically adjust the acquisition parameters and perform signal enhancement, the intervention and adjustment requirements of the operator during the experiment are greatly reduced, not only improving the automation degree of the experiment, but also reducing human errors, and high-quality signals can be obtained in a short time, thereby providing effective technical support for accurate characterization of complex reservoirs such as shale oil and gas.

[0024] As a specific embodiment, the high-speed data interface is a USB3.2 Gen2 interface, the transmission rate is ≥10 Gbps, and the delay is <0.1 ms, which meets the real-time data transmission requirements.

[0025] As a preferred embodiment, the ResNet module comprises a plurality of residual blocks, each residual block comprising at least two convolutional layers, the convolutional kernel size of each convolutional layer being arranged along the signal time domain direction, and the number of channels being increased layer by layer, for extracting the time domain features of the multi-channel signal and completing feature fusion.

[0026] As a specific embodiment, a pre-trained ResNet module is embedded in a nuclear magnetic resonance instrument signal acquisition system, connected to a host computer through a high-speed data interface (sampling rate 1MHz, transmission delay <0.1ms) and configured in real-time processing mode. The ResNet input is the first 20 echo point data of multi-channel acquisition (6 groups x 20 points, input dimension 120), which fuses multi-channel features through 3 residual blocks (each block containing 2 layers of convolution, convolution kernel size 1x3, channel number being 16, 32, 64 in turn), superimposes effective signal components, and outputs the enhanced fast relaxation component signal. Compared with the traditional single-channel method, after the ResNet fuses the multi-channel data, the signal amplitude is increased from 10μV to 30μV (increased by 200%), the signal-to-noise ratio (SNR) is increased from 15dB to 25dB, and the single processing delay is 0.72ms, which can meet the real-time requirement (sampling rate≥1MHz).

[0027] Specifically, each residual block in the ResNet module comprises two convolutional layers, the convolutional kernel size is 1x3, and the signal time domain direction is slid to extract the local time domain features; the number of channels is increased layer by layer (16→32→64), and the high-order features of the multi-channel signal are gradually fused. In addition, a shortcut connection is introduced between the input and the output to retain the effective components of the original signal and avoid gradient disappearance; the signal amplitude is amplified through the residual learning path.

[0028] As a preferred embodiment, the signal acquisition system comprises a multi-phase channel and a multi-frequency offset channel. The phases of the multi-phase channel are configured to be orthogonal to each other, and the frequency offsets of the multi-frequency offset channel are configured to be symmetrically offset; the multi-phase channel and the multi-frequency offset channel are used to synchronously acquire the fast relaxation component signal to generate a multi-dimensional initial echo data set.

[0029] In some embodiments, the signal acquisition system sets 4 phase channels (0°, 90°, 180°, 270°) and 2 frequency offset channels (±5 kHz) for parallel acquisition of fast relaxation component signals, each channel generating 1000 echo point data, forming 6 groups of initial echo data sets. The multi-channel design utilizes the diversity of phase and frequency to capture the potential characteristics of fast relaxation signals. These data are received by the ResNet module, which utilizes residual structure to fuse the effective components of multiple signals, thereby suppressing distortions introduced during acquisition due to hardware limitations (such as dead time > 0.1 ms), further enhancing signal amplitude to above 30 μV, and enhancing its strength.

[0030] As a preferred embodiment, the host computer comprises a dynamic modulation unit and a monitoring feedback unit; The dynamic modulation unit is used to dynamically adjust the radio frequency parameters of the CPME pulse sequence based on the physical properties of the sample and real-time signal feedback; The monitoring feedback unit is used to monitor the amplitude and decay characteristics of the enhanced signal in real time, and dynamically adjust the acquisition parameters and processing strategy through a feedback mechanism.

[0031] Embodiment 2 The present embodiment also provides a residual neural network-based nuclear magnetic resonance fast relaxation component signal enhancement method, which is applied to the residual neural network-based nuclear magnetic resonance fast relaxation component signal enhancement system of embodiment 1, comprising: Step S101: calibrate the initial CPME pulse parameters of the low-field nuclear magnetic resonance device; Step S102: obtain a core sample, optimize the CPME pulse sequence according to the pre-experiment results of the core sample, and determine the optimal pulse power; Step S103: perform nuclear magnetic scanning on the core sample based on the optimized CPME pulse sequence, split the signal to 6 input channels of the ResNet module through a multiplexer, and input the signal into the pre-trained ResNet module; Step S104: utilize the ResNet module to extract and fuse the time-frequency characteristics of the multi-channel signal, superimpose the effective signal components, and output the enhanced fast relaxation component signal; Step S105: dynamically adjust the CPME pulse sequence parameters according to the sample properties and real-time signal feedback, and input the modulated signal into the ResNet module for secondary enhancement; Step S106: monitor the signal amplitude and decay characteristics in real time, dynamically adjust the acquisition parameters and processing strategy in combination with the feedback control mechanism, and keep the enhanced signal amplitude stable.

[0032] The method of the embodiment can effectively improve the acquisition and processing capability of low-field nuclear magnetic resonance equipment on fast-relaxation signals, enhance signal strength, stability and adaptability, and ensure efficient and accurate core sample analysis by optimizing the CPMG pulse sequence, using multi-channel parallel acquisition and ResNet module for signal enhancement, combining dynamic parameter adjustment and real-time feedback control.

[0033] As a preferred embodiment, the setting of the initial radio frequency pulse parameters includes adjusting the radio frequency pulse power, the initial phase, and the echo interval.

[0034] As a specific embodiment, in the step S101, the calibration step includes calibrating the radio frequency coil matching degree by a standard sample, adjusting the magnetic field uniformity to within ±0.1 ppm, and ensuring the accuracy of the initial pulse parameters (power, phase, TE).

[0035] Further, in order to enhance the initial strength of the fast-relaxation component signal (T2 < 1ms), the radio frequency pulse power (such as increasing the 90° pulse to 6V and the 180° pulse to 12V) and the initial phase (such as optimizing to 45° within the range of 0°-90°) are adjusted in the CPMG pulse sequence, so that the initial amplitude of the front echo signal (the first 20 echo points) is increased from less than 10μV to more than 20μV. At the same time, the pre-trained ResNet module is initialized and embedded into the acquisition system through a high-speed data interface (sampling rate 1MHz), and is configured in real-time processing mode (delay <0.8ms) to ensure that the subsequent steps can directly use the enhanced high-quality input data.

[0036] As a preferred embodiment, in the step S102, the optimization of the CPMG pulse sequence according to the pre-experiment results of the core sample includes: gradually increasing the 90° pulse power to determine that the amplitude variation of a preset number of echo points meets the first preset standard; setting the 180° pulse power to verify that the signal decay rate meets the second preset standard; real-time monitoring of the signal waveform by an oscilloscope ensures that the decay rate is controlled within a preset range.

[0037] In some embodiments, when the 90° pulse power is increased from 4V to 6V and the amplitude of the first 20 echo points is increased from 8μV to 20μV, it is confirmed that the amplitude variation meets the requirements; the 180° pulse power is set to 12V, and the signal decay rate is reduced to 0.8ms⁻¹, which can meet the subsequent enhancement requirements.

[0038] As a preferred embodiment, in the step S104, the time-frequency feature extraction and fusion of the multi-channel signal by the ResNet module includes: The time domain feature extraction is performed on the multi-channel data, and the time-frequency characteristics of the fast relaxation signal are separated through short-time Fourier transform. The extracted feature vector is input into the ResNet module, and the residual learning path is used to jointly analyze and optimize the magnetic resonance signal data from different channels, thereby enhancing the detectability of the signal in a low signal-to-noise ratio environment.

[0039] As a specific embodiment, when performing time domain feature extraction on the multi-channel collected signal, a short-time Fourier transform (window size 5 echo points, step size 2 echo points) is used to capture the transient decay characteristics of the fast relaxation signal, and a Fourier transform is performed on the signal in each window to extract time-frequency domain features (such as amplitude spectrum, phase spectrum, and decay rate) and generate a feature vector (dimension about 20x6). These features are input into the ResNet module, and the residual learning capability of the ResNet module is used to optimize the intensity and detectability of the signal. For example, by deepening the residual path to amplify the peak amplitude (target > 35 μV), while smoothing the decay curve, the fast relaxation signal is more easily distinguished in a low signal-to-noise ratio environment.

[0040] As a preferred embodiment, in step S105, the parameters of the dynamically modulated CPMG pulse sequence include: The step value of the echo interval is adjusted according to the real-time signal amplitude, and the radio frequency pulse power is adjusted in combination with the sample porosity.

[0041] Specifically, the dynamic acquisition parameter modulation adjusts the pulse sequence parameters according to the sample characteristics and real-time signal feedback. For example, for a high porosity sample, the TE is shortened to 0.6 ms to increase the number of acquisition points; when the signal amplitude is lower than 25 μV, the pulse power is increased to 6.5 V. The modulated signal is re-input into the ResNet module, and the features are further optimized through residual learning to ensure that the enhancement effect adapts to different reservoir conditions.

[0042] As a preferred embodiment, in step S106, the feedback control mechanism includes: The amplitude and decay rate of the enhanced signal are monitored; When the amplitude is lower than a preset threshold, the number of phase channels or the frequency offset range is increased to maintain the stability of the signal; when the decay rate exceeds a preset speed, the echo interval is shortened and the ResNet input is updated synchronously.

[0043] Specifically, the enhanced fast relaxation component signal is integrated into the nuclear magnetic resonance instrument, and feedback optimization is performed by monitoring the signal amplitude and decay characteristics (such as T2 distribution peak position) in real time. If the amplitude is lower than the threshold, the system automatically adjusts the number of phase channels (such as increasing to 6) or the frequency offset range (such as ±7 kHz); if the decay is too fast (T2 < 0.5 ms), the TE is shortened to 0.5 ms and the ResNet input is updated synchronously. This dynamic optimization ensures that the signal enhancement effect remains stable when the sample is replaced (such as shale to sandstone) or the environment changes (such as temperature fluctuations ±5℃).

[0044] In order to better illustrate the method of the present application, the actual operation process is shown below through a specific example.

[0045] As a specific example, first, the experimental environment and instruments need to be configured; specifically, a low-field nuclear magnetic resonance instrument with a working frequency of 2MHz (magnetic field strength 0.05T) is selected, and an experimental platform is built in a constant temperature laboratory (temperature controlled at 25℃, humidity 50%). The radio frequency pulse parameters of the CPMG pulse sequence of the instrument are calibrated, the 90° pulse width is set to 10μs (power 6V), the 180° pulse width is set to 20μs (power 12V), and the initial signal amplitude is verified to be above 15μV through a standard sample, ensuring that the instrument is in a stable working state.

[0046] Second step: sample preparation and characteristic recording; cylindrical core samples (diameter 2.5cm, height 5cm) are extracted from shale oil and gas reservoirs, and the physical characteristics of the samples are recorded. The sample surface is wiped clean with laboratory paper, then placed in a vacuum drying oven (temperature 120℃, vacuum degree 0.01MPa) for 48 hours to ensure that the sample is in a water-free and oil-free state.

[0047] Third step: collect the initial fast relaxation signal: configure the improved CPMG pulse sequence, set the initial echo interval (TE) to 0.8ms, the repetition time (TR) to 1.5s, the echo point number to 1000, and the radio frequency pulse initial phase to 45° (based on phase scanning optimization).

[0048] Determine the optimal pulse power through pre-experiment: when the 90° pulse power increases from 4V to 6V, the amplitude of the first 20 echo points increases from 8μV to 20μV (see Table 1); when the 180° pulse power is set to 12V, the signal decay rate is reduced to 0.8ms⁻¹, meeting the subsequent enhancement requirements.

[0049] Table 1: Performance comparison between traditional method and the present application It should be noted that during the acquisition process, the oscilloscope (bandwidth 1 MHz, sampling rate 2 MS / s) was used to monitor the signal waveform in real time, and the decay rate was controlled within 0.9 ms⁻¹ to provide reliable raw data for subsequent enhancement.

[0050] Fourth step: multi-channel signal acquisition and fusion preparation: multi-channel parallel acquisition was implemented, 4 phase channels (0°, 90°, 180°, 270°, phase accuracy ±2°) and 2 frequency offset channels (+5 kHz and -5 kHz, offset accuracy ±0.5 kHz, bandwidth 10 kHz±1 kHz) were configured, and 1000 echo points of data were synchronously acquired for each channel, generating 6 groups of initial echo data sets (data format 16-bit integer, total size about 120 kB). The signal distributor (bandwidth >50 kHz, channel isolation >40 dB) was used to distribute the collected signals to the multi-channel input port, ensuring that the cross-talk between channels was lower than -50 dB. The data was stored in the buffer (capacity >1 MB, read / write speed >10 MB / s) to prepare for multi-dimensional input for ResNet module fusion processing.

[0051] Fifth step: ResNet module embedding and signal enhancement processing: the pre-trained ResNet module was embedded in the magnetic resonance instrument signal acquisition system, connected to the host computer through a high-speed data interface (sampling rate 1 MHz, transmission delay <0.1 ms), and configured in real-time processing mode. The ResNet input was the first 20 echo point data of multi-channel acquisition (6 groups x 20 points, input dimension 120), which fused multi-channel features through 3 residual blocks (each block containing 2 layers of convolution, convolution kernel size 1 x 3, channel number 16, 32, 64 in turn), superimposed effective signal components, and output the enhanced fast relaxation component signal. Compared with the traditional single-channel method, after ResNet fused multi-channel data, the signal amplitude was increased from 10 μV to 30 μV (increased by 200%), the signal-to-noise ratio (SNR) was increased from 15 dB to 25 dB, and the single processing delay was 0.72 ms (see Table 1), which met the real-time requirement (sampling rate ≥1 MHz). Sixth step: dynamic parameter modulation and enhanced signal output: according to the sample characteristics and real-time signal feedback, the CPMG pulse sequence parameters were dynamically adjusted: TE range 0.6 ms-1.0 ms (step 0.02 ms), pulse amplitude range 4 V-6.5 V (step 0.1 V). For example, when the amplitude is lower than 25 μV, the TE is shortened to 0.6 ms and the power is increased to 6.5 V, and the number of acquisition points is increased to 12 per millisecond.

[0052] The modulated signal is input into the ResNet module for secondary enhancement, and the output signal is integrated into the instrument display system (resolution 16 bits, refresh rate > 10 Hz), and the number of phase channels is adjusted in real time through the feedback control unit (processing frequency > 100 MHz) to ensure that the enhanced signal amplitude is stable above 30 mu V.

[0053] As shown in Figure 2 , Figure 2 The method flow diagram of the embodiment is shown.

[0054] The system and method for enhancing the signal of the fast-relaxation component based on the residual neural network provided by the application amplify the front echo amplitude of the fast-relaxation signal in real time through residual learning, and the initial signal amplitude is increased from 10 mu V to 30 mu V in a low-field device; the detectability of the enhanced signal in a low signal-to-noise ratio environment is combined with the multi-phase and multi-frequency channel data and ResNet; a self-adaptive fast-relaxation signal enhancement mechanism is formed through the dynamic modulation of the acquisition parameters and the cooperative optimization of ResNet. The application uses the residual learning ability of ResNet to enhance the fast-relaxation component signal from two aspects of acquisition optimization and data processing, and provides technical support for the accurate characterization of complex reservoirs such as shale oil and gas.

[0055] The above is only a preferred specific embodiment of the application, but the protection scope of the application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the application, which should be covered within the protection scope of the application.

Claims

1. A nuclear magnetic resonance fast relaxation component signal enhancement system based on residual neural network, characterized in that: The invention comprises a low-field nuclear magnetic resonance device and a main control computer connected to each other; wherein the low-field nuclear magnetic resonance device includes a signal acquisition system, which is connected to the main control computer via a high-speed data interface and is embedded with a pre-trained ResNet module; The low-field nuclear magnetic resonance device is configured with a CPMG pulse to excite a sample under test to generate a fast relaxation signal. Multi-dimensional echo data is collected in parallel using multiple channels, and feature fusion is performed on the multi-channel data using a ResNet module to enhance signal amplitude, which is then transmitted to a main control computer. The ResNet module is pre-trained on a fast relaxation signal dataset containing typical reservoir samples; the dataset contains multi-channel echo data of different porosities and fluid types. The main control computer is configured in real-time processing mode to obtain real-time signals from the signal acquisition system and dynamically adjust acquisition parameters and processing strategies based on the physical properties of the sample being tested and the feedback control mechanism to maintain signal stability.

2. The nuclear magnetic resonance fast relaxation component signal enhancement system based on residual neural network according to claim 1, characterized in that The ResNet module includes multiple residual blocks, each residual block includes at least two convolutional layers, the convolution kernel size of each convolutional layer is set along the signal time domain direction, and the number of channels of each convolutional layer increases by 2 times layer by layer.

3. The nuclear magnetic resonance fast relaxation component signal enhancement system based on residual neural network according to claim 1, characterized in that: The signal acquisition system includes multiple phase channels and multiple frequency offset channels; The phases of the multi-phase channels are configured to be orthogonal to each other, and the frequency offsets of the multi-frequency offset channels are configured to be symmetrically offset; the multi-phase channels and the multi-frequency offset channels are used to synchronously acquire fast relaxation component signals to generate a multidimensional initial echo data set.

4. The nuclear magnetic resonance fast relaxation component signal enhancement system based on residual neural network according to claim 1, characterized in that: The main control computer includes a dynamic modulation unit and a monitoring feedback unit; The dynamic modulation unit is used to dynamically adjust the radio frequency parameters of the CPMG pulse sequence based on the physical characteristics of the sample under test and real-time signal feedback; The monitoring feedback unit is used to monitor the amplitude and attenuation characteristics of the enhanced signal in real time, and dynamically adjust the acquisition parameters and processing strategy through a feedback mechanism.

5. A method for enhancing the signal of a fast-relaxing component of nuclear magnetic resonance based on a residual neural network, applied to the system for enhancing the signal of a fast-relaxing component of nuclear magnetic resonance based on a residual neural network as claimed in any one of claims 1 to 4, characterized in that: include: Calibrate the initial CPMG pulse parameters of low-field NMR equipment; Obtain core samples, optimize the CPMG pulse sequence based on the preliminary experimental results of the core samples, and determine the optimal pulse power; Based on the optimized CPMG pulse sequence, the core samples were scanned by nuclear magnetic resonance (NMR), and the fast relaxation component signals were collected in parallel through multiple channels. The signals were then split and input into the pre-trained ResNet module. The ResNet module is used to extract and fuse the time-frequency features of multi-channel signals, superimpose the effective signal components, and output the enhanced fast relaxation component signal; According to the amplitude and decay rate of the fast relaxation component signal, the pulse power and echo interval parameters of the CPMG pulse sequence are dynamically adjusted, and the modulated signal is input into the ResNet module for secondary enhancement; The amplitude and attenuation characteristics of the enhanced signal are monitored in real time, and the acquisition parameters and processing strategies are dynamically adjusted in combination with the feedback control mechanism to keep the enhanced signal amplitude stable.

6. The method for enhancing the fast relaxation component signal of nuclear magnetic resonance based on residual neural network according to claim 5, wherein: The initial CPMG pulse parameters include: radio frequency pulse power, initial phase and echo interval.

7. The method for enhancing the fast relaxation component signal of nuclear magnetic resonance based on residual neural network according to claim 5, wherein: The method of using the ResNet module to extract and fuse time-frequency features of multi-channel signals includes: Perform time domain feature extraction on multi-channel data and separate the time-frequency features of fast relaxation signals through short-time Fourier transform; The extracted feature vectors are input into the ResNet module, and the residual learning path is used to jointly analyze and optimize the nuclear magnetic resonance signal data from different channels to enhance the detectability of the signal in a low signal-to-noise ratio environment.

8. The method for enhancing the fast relaxation component signal of nuclear magnetic resonance based on residual neural network according to claim 5, wherein: The parameters of the dynamically modulated CPMG pulse sequence include: The step value of the echo interval is adjusted according to the real-time signal amplitude, and the power of the CPMG pulse sequence is adjusted in combination with the porosity of the sample being tested.

9. The method for enhancing the fast relaxation component signal of nuclear magnetic resonance based on residual neural network according to claim 5, wherein: The feedback control mechanism includes: Monitor the amplitude and decay rate of the enhanced signal; When the amplitude is lower than the preset threshold, the number of phase channels is increased or the frequency offset range is expanded to maintain signal stability; when the attenuation speed exceeds the preset speed, the echo interval is shortened and the ResNet input is updated synchronously.

10. The method for enhancing the fast relaxation component signal of nuclear magnetic resonance based on residual neural network according to claim 5, characterized in that: The optimization of the CPMG pulse sequence according to the preliminary experimental results of the core sample includes: gradually increasing the 90° pulse power to determine that the amplitude changes of a preset number of echo points meet a first preset standard; Setting the 180° pulse power and verifying that the 180° pulse signal attenuation rate meets a second preset standard; The signal waveform is monitored in real time through an oscilloscope to ensure that the attenuation rate is controlled within the preset range.

Images