A method and system for real-time detecting the accumulation state of hydrates in a core sample

Abstract

The present invention relates to the field of geological exploration, and aims to provide a method and a system for real-time detection of the hydrate accumulation state of a core sample. The method and system include: using a sample displacement device to continuously move a cylindrical core sample axially through the sample channel of a sound velocity detection device; during this process, the sound velocity detection device emits pulsed acoustic waves along the radial direction of the core sample and receives the acoustic wave signals after penetrating the core sample; after a data acquisition card acquires the acoustic wave signals, the signals are sent to a computer for processing to draw a real-time waveform diagram; the data of the real-time waveform diagram is subjected to denoising and normalization processing to form real-time input data of the core sample; the input data of the core sample is input into a trained neural network, and a real-time prediction result is obtained through calculation. Without damaging the core sample, the present invention can quantitatively predict the hydrate accumulation state, with higher accuracy and faster data processing speed; it can greatly improve the work efficiency, provide an analysis basis for measurement, and further expand the application scenarios.

Description

Technical Field

[0001] The present invention relates to the field of geological exploration, and particularly to a method and a system for real-time detecting the accumulation state of hydrates in core samples. Background Art

[0002] Petrophysical experiments have always played an important role in the research of the geophysical field. Their experimental results are usually considered to be the most accurate, and thus a large number of petrophysics studies are carried out based on this. However, petrophysical experiments often cause damage to the core and are difficult to reuse, which to a certain extent increases the research cost or causes discontinuity of research subjects.

[0003] The elastic mechanical parameters of rocks (such as bulk modulus, shear modulus, and Poisson's ratio, etc.) have an inevitable connection with their structural composition, formation and evolution, etc., and have a unique position in the research of the earth science field. The pore structure of rocks affects the acoustic properties of rocks. Measuring the radial acoustic characteristics of the core can better understand the radial damage characteristics of the core, and determining the axial acoustic characteristics of the rock can largely analyze and determine physical parameters such as the pore structure and its distribution state of the rock, and the properties of pore-filling fluids.

[0004] Chinese Patent Application CN103267802A discloses an acoustic wave detection device for natural gas hydrate cores, which is used for clamping and measuring the radial sound velocity of natural gas hydrate cores. However, this technical solution can only store and export data through an oscilloscope, and the sound velocity needs to be manually calculated after exporting the data; more importantly, this technical solution can only roughly estimate the possibility of the existence of hydrates in the core sample by the waveform change situation by technicians according to experience, and cannot make a qualitative analysis and judgment on the accumulation state of hydrates in combination with the measurement situation.

[0005] Therefore, it is necessary to propose an analysis technology for the accumulation state of hydrates in core samples with high automation, without manual intervention, and capable of directly giving quantitative analysis results. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a method and a system for real-time detecting the accumulation state of hydrates in core samples.

[0007] To solve the technical problem, the solution of the present invention is:

[0008] Provide a method for real-time detecting the accumulation state of hydrates in core samples, including the following steps:

[0009] (1) Use the sample displacement device to continuously pass the cylindrical core sample axially through the sample channel of the sound velocity detection device; during this process, the sound velocity detection device emits pulsed sound waves radially along the core sample and receives the sound wave signals after penetrating the core sample;

[0010] (2) After the data acquisition card collects the sound wave signals, it sends them to the computer for processing and draws a real-time waveform diagram;

[0011] (3) Denoise and standardize the data of the real-time waveform diagram to form real-time core sample input data;

[0012] (4) Input the core sample input data into the trained neural network, and obtain the real-time prediction results through calculation. The results include the accumulation position of the hydrate center in the cross-section of the core column, the fitting radius of the hydrate, and the sound velocity of the core.

[0013] The present invention further provides a system for real-time detecting the accumulation state of hydrates in a core sample, including a sound velocity detection device, a sample displacement device, a pulse generator / receiver, a data acquisition card, and a computer; wherein,

[0014] The sound velocity detection device includes a main body member, a transmitting ultrasonic probe, and a receiving ultrasonic probe; the main body member has a horizontally arranged cylindrical sample channel, and the transmitting ultrasonic probe and the receiving ultrasonic probe are relatively installed on the upper and lower surfaces of the main body member, and the connecting line of the centers of the two coincides with the maximum diameter direction of the sample channel;

[0015] The sample displacement device is used to pass the core sample through the sample channel of the sound velocity detection device at a set speed;

[0016] The pulse generator / receiver is respectively connected to the transmitting ultrasonic probe and the receiving ultrasonic probe through signal lines, and is used to trigger the transmitting ultrasonic probe to emit pulsed sound waves radially along the core sample and receive the sound wave signals after penetrating the core sample;

[0017] The data acquisition card is respectively connected to the pulse generator / receiver and the computer through signal lines, and is used to collect the sound wave signals received by the pulse generator / receiver and send them to the computer for processing;

[0018] The computer has built-in software function modules, which are used to draw a real-time waveform diagram and perform denoising and standardization processing to obtain the core sample input data, input it into the trained neural network, calculate the prediction results of the hydrate accumulation state, and display them on the display interface of the software function module.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. Without damaging the core sample, the present invention can quantitatively predict the hydrate accumulation state, with higher accuracy compared to the existing technology based on manual judgment methods.

[0021] 2. The present invention uses a computer to achieve data processing and neural network calculations, with fast data processing speed and high accuracy.

[0022] 3. The present invention uses a sample displacement device to pass the core sample through the sound velocity detection device at a set speed, eliminating the need for manual intermittent displacement in traditional measurement schemes, greatly improving work efficiency and solving the problem of low automation in sound velocity detection during core exploration.

[0023] 4. The present invention can provide an analysis basis for the measurement of physical parameters such as core hydrate saturation and porosity, further expanding the application scenarios of core hydrate measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a schematic diagram of the system of the present invention;

[0025] Figure 2 is a schematic diagram of the sound velocity detection device of the present invention;

[0026] Figure 3 is the longitudinal sectional view of the middle of the sound velocity detection device;

[0027] Figure 4 is the process of data standardization of the waveform diagram into an extreme value diagram;

[0028] Figure 5 is a schematic diagram of the structure of the neural network in the present invention;

[0029] Figure 6 is the training process diagram of the neural network in the present invention;

[0030] Figure 7 is the mean square error discrimination process of the neural network in the present invention;

[0031] Figure 8 is the detection flow chart of the present invention.

[0032] In the figure, 1 is the sound velocity detection device, 1-1 is the fastening stud, 1-2 is the probe pressure bridge, 1-3 is the probe pressure cap, 1-4 is the transmitting ultrasonic probe, 1-5 is the fixing stud, 1-6 is the main body member, 1-7 is the receiving ultrasonic probe; 2 is the pulse generator / receiver, 3 is the data acquisition card, 4 is the computer. DETAILED DESCRIPTION OF THE INVENTION

[0033] First of all, it should be noted that the present invention relates to big data and neural network learning, which is an application of computer technology in the field of measurement. During the implementation of the present invention, the application of multiple software functional modules will be involved. The applicant believes that after carefully reading the application documents and accurately understanding the implementation principle and the purpose of the present invention, and in combination with the existing well-known technologies, those skilled in the art can fully implement the present invention by using their mastered software programming skills. All that mentioned in the application documents of the present invention fall within this category, and the applicant will not list them one by one.

[0034] Those skilled in the art know that in addition to implementing a part of the system provided by the present invention and its various devices, modules, and units in the form of pure computer-readable program codes, the method steps can be logically programmed to enable the system provided by the present invention and its various devices, modules, and units to be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers, etc. to achieve the same functions. Therefore, the system provided by the present invention and its various devices, modules, and units can be considered as a kind of hardware component, and the devices, modules, and units included therein for implementing various functions can also be regarded as the structures within the hardware component; the devices, modules, and units for implementing various functions can also be regarded as both software modules for implementing the method and the structures within the hardware component.

[0035] It should also be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of another identical element in the process, method, article or device comprising the said element.

[0036] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only partial embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0037] The system for real-time detecting the hydrate accumulation state of a core sample in the present invention is as Figure 1As shown in the figure, it includes a sound velocity detection device 1, a specimen displacement device (not shown in the figure), a pulse generator / receiver 2, a data acquisition card 3, and a computer 4. Among them, the sound velocity detection device 1 is used to emit and receive sound waves to the core, and then detect the sound velocity of the core; the pulse generator / receiver 2 is used for the generation of pulse electrical signals and the reception of acoustic electrical signals; the data acquisition card 3 is used for the acquisition of the received acoustic signals; the computer 4 is used for the processing of the acoustic signals. Among them,

[0038] The sound velocity detection device 1 includes a main body member, a transmitting ultrasonic probe 1-4, and a receiving ultrasonic probe 1-7; the main body member has a horizontally arranged cylindrical specimen channel, and the transmitting ultrasonic probe 1-4 and the receiving ultrasonic probe 1-7 are relatively installed on the upper and lower surfaces of the main body member, and the connection line of the centers of the two coincides with the maximum diameter direction of the specimen channel. Both probes are piezoelectric crystal longitudinal wave probes, which are used to realize the mutual conversion between mechanical acoustic signals and electrical signals. The specimen displacement device is used to pass the core specimen through the specimen channel of the sound velocity detection device 1 at a set speed. The pulse generator / receiver 2 is respectively connected to the transmitting ultrasonic probe 1-4 and the receiving ultrasonic probe 1-7 through signal lines, and is used to trigger the transmitting ultrasonic probe 1-4 to emit pulse sound waves along the radial direction of the core specimen and receive the sound wave signals after penetrating the core specimen. The data acquisition card 3 is respectively connected to the pulse generator / receiver 2 and the computer 4 through signal lines, and is used to acquire the acoustic signals received by the pulse generator / receiver 2 and send them to the computer 4 for processing. The computer 4 has built-in software function modules, which are used to draw real-time waveform diagrams and perform denoising and normalization processing to obtain test sample data, input it into a trained neural network, calculate the prediction result of the hydrate accumulation state through calculation, and display it on the display interface of the software function module.

[0039] As Figure 2As shown, the main component 1-6 of the sound velocity detection device 1 is made of 316 stainless steel, and the outer cross-section is a square with a side length of 140mm, which is used to fix the transmitting ultrasonic probe 1-4 and the receiving ultrasonic probe 1-7. The sample channel of the main component 1-6 is used to carry the core sample, and its cross-section is a circle with a diameter of 60mm. The upper and lower outer surfaces of the main component 1-6 each have a circular groove with a diameter of 32mm and a depth of 3.7mm, which is used to form a match with the ultrasonic probe. Near the circular groove, there are two stud holes with a diameter of 8mm distributed along the longitudinal section of the main component 1-6, which are used to be fixed with the holes on the bridge foot of the probe pressure bridge 1-2 through studs. The two ultrasonic probes are cylindrical, with signal line interfaces on the side, and are connected to the pulse generator / receiver 2 through a BNC line. A circular boss is set on the bottom surface of the probe, and the surface of the boss is used as the ultrasonic contact surface, which is pressed tightly into the circular concave on the surface of the main component 1-6 and connected to the pulse generator / receiver 2 through a BNC line. The probe pressure bridge 1-2 is made of 316 stainless steel, with a bridge-shaped cross section. The bridge foot has two 8mm circular through holes, which are fixed to the main component 1-6 by fixing studs 1-5. There is an 8mm circular through hole in the middle of the probe pressure bridge 1-2, and the transmitting ultrasonic probe 1-4 and the receiving ultrasonic probe 1-5 are pressed by the fastening studs 1-1. The probe pressure cap 1-3 is a circular sheet with a lower edge, and the upper part is fixed with a nut-shaped clamp block with a hexagonal appearance and internal threads. It covers the top of the transmitting ultrasonic probe 1-4 and the receiving ultrasonic probe 1-, and cooperates with the fastening studs in the center of the probe pressure bridge 1-2 to disperse the stress on the probe.

[0040] The longitudinal section of the sound velocity detection device is as follows Figure 3 As shown, the sound wave is emitted by the transmitting ultrasonic probe 1-4, propagates along the diameter direction of the core sample, and is received by the receiving ultrasonic probe 1-7.

[0041] The process of converting the waveform graph into an extreme value graph through data standardization is as follows: Figure 4 As shown, after the data acquisition card 3 obtains the real-time sound wave graph, it denoises the initial sample data set to obtain a denoised waveform. Specifically, it includes:

[0042] By reading the time coordinate t and amplitude m of the extreme point in the real-time waveform diagram, combined with the sample diameter d, the initial sample data set is formed; then the obtained test waveform is subjected to multi-layer wavelet decomposition (wavedec function) to obtain the corresponding detail coefficient and approximate coefficient, that is, according to the wavelet principle, the detail coefficient and approximate coefficient of the corresponding layer are set to zero, and the denoised waveform is reconstructed to eliminate abnormal values. Finally, the detail coefficient is used to find the maximum and minimum points by marking the slope, and the data is normalized to calibrate the time coordinate t of the extreme point. i and amplitude m i , and get the data normalization graph. Among them, t i 、mi is the coordinate array of all extreme points in this waveform, t i = {t i1 , t i2 , t i3 …}, m i = {m i1 , m i2 , m i3 …}, and the current sample number is represented by i.

[0043] The structure of the neural network is as Figure 5 shown. The neural network is an artificial neural network (ANN) constructed based on a multi-layer neural network (MLP), consisting of an input layer, a hidden layer, and an output layer, and all are taken as one layer; each layer contains multiple neurons, and the neurons in each layer are connected to the neurons in the adjacent layer; considering minimizing the training time while ensuring the prediction accuracy, the hidden layer is taken as one layer. Among them,

[0044] The input layer takes the time coordinate t i of the extreme points in the waveform diagram and the radial distance d i of the core sample as inputs;

[0045] The number of neurons in the hidden layer is determined by the empirical formula (1):

[0046]

[0047] where: N H , N I , N O are the number of neurons in the hidden layer, the input layer, and the output layer respectively, and N T represents the number of training samples.

[0048] The output parameters of the output layer are the sound velocity C out in the cross-section where the center plane of the ultrasonic probe of the core column is located, the accumulation position (x out , y out ) of the predicted hydrate center in the core column, and the fitting radius R out of the hydrate. Among them, x out represents the horizontal (lateral) offset of the hydrate accumulation point relative to the reference base point at the cross-section of the cylindrical surface where the center point of the acoustic probe is located, y out represents the vertical (longitudinal) offset of the hydrate accumulation point relative to the reference base point, and R out represents the radius after fitting the irregular hydrate accumulation range into a nearly circle according to the predicted hydrate accumulation position.

[0049] For the BP neural network in the figure, the size of the input layer X is N*D, where N is the time coordinate t i and the amplitude mi , and the core diameter d of the core column i , D is the number of samples, W1 and W2 are the weight matrices of the hidden layer and the output layer respectively, B1 and B2 are the biases of the hidden layer and the output layer respectively, and the output Y is the hydrate accumulation position (x out , y out ), the fitting radius R of the hydrate out , and the sound velocity C out . Each neuron obtains an output value by weighting all its inputs and passing through an activation function. The weights of the input data are the parameters of the neural network. Since there will be an error between the output value and the target value, the neural network needs to be trained with a sample data set, and the weights W and biases B are continuously optimized during this process.

[0050] As Figure 6 shown in the training process diagram, after the neural network is constructed, 500 standard core columns containing natural gas hydrate are artificially made. Taking the radial center of the standard core column as the reference base point, the contact point of the acoustic probe and the core column is denoted as A coordinate system is established, and different positions (x1, y1), (x2, y2), (x3,...), (x n , y n ) of different cross-sections in the core column are used as the centers, and spheroid natural gas hydrates with similar volume sizes of fitting radii R1, R2, R3,... are built in. Using this system, multiple acoustic tests are carried out on the core column with a radial distance of d n and the center position of the natural gas hydrate being (x i , y i , y i ). A set of test waveforms of each core column is obtained, and the average center position (x i , y i ) of the hydrate in the adjacent length cross-section of the core column where the hydrate is stored and the fitting radius R of the hydrate are obtained through physical means of artificial measurement i, after processing all the obtained acoustic wave signal diagrams, they are used as the initial sample data set of the neural network model. Further, the process of obtaining sample data is as follows: Use this system to conduct a test on a certain position of a core column, and manually measure the sound velocity C1, the core cross-section diameter d1, the accumulation position (x1, y1) of the hydrate center in the core, and the fitting radius R1 of the hydrate in the cross-section of the core where the center plane of the ultrasonic probe is located. At this position, the time coordinates t1 and amplitudes m1 of different extreme points after the acoustic wave signal passes through the waveform extreme points are determined. Then, taking t1, m1, and d1 as input parameters and C1, (x1, y1), and R1 as the expected output parameters (true values), it is a set of training data. Then measure 500 cores, conduct a test at a certain distance Δs intervals on the core column, measure 10 groups of data for each core, and obtain a total of 5000 groups of data to form a learning sample set. Let i represent the current sample serial number, and the number of learning times should be less than the number of samples.

[0051] Denote the denoised and normalized sample data set (5000) as D s , and denote D s as being divided into a training set (90%, 4500) and a test set (10%, 500), where the waveforms in the training set should have both samples containing hydrates and test samples without hydrates;

[0052] Train the constructed neural network. The main learning method is the backpropagation learning method of the artificial neural network. By comparing the difference between the output result and the sample expectation, and feeding back this difference to modify the weights of the neurons in the hidden layer (in the neural network, it represents the association strength between two neurons. A positive weight value indicates that the front-end neuron plays an excitatory role (positive correlation) on the back-end neuron, and a negative value indicates that the front-end neuron plays an inhibitory role (negative correlation)) parameter w to reduce the error. This process will be repeated until the number of training times reaches the pre-designed upper limit or the mean square error of the network structure drops to meet the requirements.

[0053] The calculation method of the mean square error is determined by formula (2):

[0054]

[0055] In the formula: ΔN represents the mean square distance difference between the output value and the target value, that is, the mean square error; represents the expected output, represents the current training output; i is the current training sample serial number, and j is the dimension of the training sample data.

[0056] Set the upper limit of training to 4000 times and set the critical mean square error to 0.05. The optimization algorithm uses the stochastic gradient descent method and uses the rectified linear unit function f(x k; θ) as the activation function, the learning rate determines the amount of weight change in each cycle of training and is set to 0.1 to ensure the stability and convergence efficiency of the system.

[0057] The loss function loss(θ) of the neural network is the square L 2 norm, which is determined by formula (3):

[0058]

[0059] In the formula: N T represents the number of training samples; x k represents the input of the neuron at the kth observation value; θ is the input setting of the neural network, θ = {w 1 , b 1 , w 2 , b 2 ,..., w s , b s}, w represents the weight, b represents the threshold value in cooperation with the rectified linear unit; s represents the layer number; y represents the true value of the sample; y i represents the true value of the ith sample; F(x ki ; θ) represents the predicted value of the ith sample; x ki represents the input of the neuron at the ith sample.

[0060] For the specific setting of θ, please refer to Figure 7 . It includes: using the regression function from the hidden layer to the output layer to calculate the output of each hidden layer unit. The deviation is defined as the mean square distance difference ΔN between the output value and the target value. If it is within the allowable range, the training ends, and the input θ of the threshold and weight parameters is determined. Otherwise, the error is backpropagated, multiplied by the weights of the nodes connected to the output layer to obtain the error of each node, and the bias value is added to the original threshold and weight to update θ, and then re-enter the iterative calculation until the evaluation requirements are met and the training of the network is completed.

[0061] For the output results (x out , y out ) and R out , the coefficient of determination R 2 and the error function L are used as evaluation indicators, and it is judged whether they meet the credibility requirements.

[0062] Specifically, using the vertical offset y out of the hydrate accumulation point relative to the reference base point, the coefficient of determination R 2 and the error function L are obtained based on the calculations of formulas (4) and (5):

[0063]

[0064] In the formula, ya is the true value, y out represents the predicted value, y a represents the average value; when the R 2 value is 90% or above, the smaller the L, the better the prediction effect of the neural network and the more credible the output;

[0065] In the same way, the horizontal offset x of the hydrate accumulation point out is evaluated;

[0066] When the coefficient of determination R 2 meets the requirements, it indicates that the accuracy of the predicted position is guaranteed, and the error from the expected output value is within the allowable range. Therefore, the output (x out , y out ), the fitting radius R of the hydrate out is credible.

[0067] As Figure 8 shown, the method for real-time detecting the hydrate accumulation state in the core sample in the present invention includes the following steps:

[0068] (1) System connection. Connect the through interface, receiving interface, receiving output interface of the pulse generator / receiver 2, the synchronization interface, and the USB interface of the data acquisition card 3 to the transmitting ultrasonic probe 1-4, the receiving ultrasonic probe 1-7, the synchronization signal interface of the data acquisition card 3, the acoustic wave signal interface, and the computer 4 respectively through cables.

[0069] (2) Use the sample displacement device to continuously pass the cylindrical core sample axially through the sample channel of the acoustic velocity detection device 1; during this process, the acoustic velocity detection device 1 emits pulsed acoustic waves radially along the core sample and receives the acoustic wave signals after penetrating the core sample;

[0070] (3) After the data acquisition card acquires the acoustic wave signals, send them to the computer and use the MATLAB software to draw a real-time waveform diagram. First, use the data acquisition card 3 to receive the acoustic wave electrical signals, and the sampling rate is 250M s -1 . Use the MATLAB software to draw a sampling point curve diagram with the number of sampling points as the abscissa and the digital voltage as the ordinate, that is, the real-time waveform diagram. In the waveform diagram, multiple waveforms will respond after each pulse signal.

[0071] (4) After obtaining the real-time waveform diagram, enter the data processing process and perform denoising and normalization processing. Read the coordinate values t input , m input of the extreme points in the signal diagram, and combine with the diameter d input of the test core to form real-time test sample data.

[0072] (5) Import the real-time test sample data into the trained neural network. After calculation, the prediction results are obtained, which include the accumulation position (x out , y out ) of the hydrate center in the core column and the fitting radius R out of the hydrate.

[0073] (6) In the display interface of the computer, according to the calculated accumulation position of the hydrate center and the fitting radius of the hydrate, the simulation diagram of the predicted position of the hydrate in the specimen is displayed in real time.

Claims

1. A method for real-time detecting the accumulation state of hydrate in a core sample, characterized in that, It includes the following steps: (1) Use the specimen displacement device to continuously pass the cylindrical core specimen axially through the specimen channel of the sound velocity detection device; during this process, the sound velocity detection device emits pulsed acoustic waves along the radial direction of the core specimen and receives the acoustic wave signals after penetrating the core specimen; (2) After the data acquisition card acquires the acoustic wave signals, it sends them to the computer for processing and draws a real-time waveform diagram; (3) Denoise and standardize the data of the real-time waveform diagram to form real-time core sample input data; (4) Input the core sample input data into the trained neural network, and obtain the real-time prediction result through calculation. The result includes the accumulation position of the hydrate center in the cross-section of the core column, the fitting radius of the hydrate, and the sound velocity of the core.

2. The method according to claim 1, wherein The specific content of step (3) includes: (3.1) Read the time coordinates t and amplitudes m of the extreme points in the real-time waveform diagram, and combine with the specimen diameter d to form an initial sample data set; (3.2) Denoise the initial sample data set: First, use the wavedec function to perform multi-layer wavelet decomposition on the real-time waveform to obtain the corresponding detail coefficients and approximate coefficients; according to the wavelet principle, set the detail coefficients and approximate coefficients of the corresponding layers to zero, complete the reconstruction of the denoised waveform, and eliminate outliers; finally, use the detail coefficient to find the maximum and minimum points by marking the slope and perform standardization; calibrate the time coordinates of the extreme points t i and amplitude m i , get a standardized sample data set for input into the neural network.

3. The method according to claim 1, characterized in that, The neural network is an artificial neural network ANN constructed based on the multi-layer neural network MLP, which consists of an input layer, a hidden layer, and an output layer, and all are taken as one layer; each layer contains multiple neurons, and the neurons of each layer are connected to the neurons of the adjacent layer; among them, The input layer uses the time coordinate t of the extreme points in the waveform diagram i and the radial distance d of the core sample i as inputs; The number of neurons in the hidden layer is determined by the empirical formula (1): Where: N H , N I , N O are the numbers of neurons in the hidden layer, input layer, and output layer respectively, and N T represents the number of training samples; The output parameters of the output layer include the sound velocity C in the radial direction of the core sample out , as well as the predicted accumulation position (x out , y out ) of the hydrate center and the fitting radius R out of the hydrate; where x out represents the horizontal offset of the hydrate accumulation point relative to the reference base point, and y out represents the vertical offset of the hydrate accumulation point relative to the reference base point; the hydrate fitting radius R out refers to the radius after fitting the irregular hydrate accumulation range into an approximate circle according to the predicted accumulation position of the hydrate.

4. The method according to claim 1, characterized in that, The specific content of step (4) includes: Train the neural network with the backpropagation learning method of the artificial neural network: by comparing the difference between the output result and the sample expectation, and feedback this difference to modify the weight parameter w of the hidden layer neurons to reduce the error; repeat this process until the number of training times reaches the preset upper limit, or the mean square error of the neural network drops to meet the requirements; Among them, the calculation method of the mean square error is determined by formula (2): Where: ΔN represents the mean square distance difference between the output value and the target value, that is, the mean square error; T i j represents the expected output, O i j represents the current training output; i is the current training sample serial number, and j is the training sample data dimension; In the neural network, the optimization algorithm adopts the stochastic gradient descent method, and uses the rectified linear unit function f(x k ; θ) as the activation function; the learning rate is set to 0.1 to ensure the stability and convergence efficiency of the system; the loss function loss(θ) is the squared L 2 norm, which is determined by formula (3): Where: N T represents the number of training samples; x k represents the input of the neuron at the k-th observation; θ is the input setting of the neural network, θ = {w 1 , b 1 , w 2 , b 2 ,..., w s , b s}, w represents the weight, b represents the threshold cooperating with the rectified linear unit; s represents the layer number; y represents the true value of the sample; y i represents the true value of the i-th sample; F(x ki ; θ) represents the predicted value of the current sample; x ki represents the input of the neuron at the i-th sample; The specific setting method of θ includes: use the regression function from the hidden layer to the output layer to calculate the output of each hidden layer unit; if the mean square distance difference ΔN is within the allowable range, the training ends, and the input θ of the threshold and weight parameters is determined; otherwise, the error is propagated back, multiply the error by the weight of the node connected to the output layer to obtain the error of each node, and add a bias amount to the original threshold and weight to update θ, and re-enter the iterative calculation until the evaluation requirements are met and the training of the network is completed; For the output result (x out , y out ), and R out , the coefficient of determination R 2 and the error function L are used as evaluation indicators, and it is determined whether the credibility requirement is met.

5. The method according to claim 4, wherein Using the vertical offset y of the hydrate accumulation point relative to the reference base point out , the coefficient of determination R is obtained based on the calculations of formulas (4) and (5) 2 and the error function L: where y a is the true value, y out represents the predicted value, and y a represents the average value; N T represents the number of training samples; when the R 2 value is 90% or above, the smaller the L, the better the prediction effect of the neural network and the more credible the output; In the same way, the horizontal offset x of the hydrate accumulation point is evaluated; out ​ When the coefficient of determination R 2 meets the requirements, it indicates that the accuracy of the predicted position is guaranteed, and the error from the expected output value is within the allowable range. Therefore, the output of (x out , y out ), and the fitting radius R out of the hydrate are credible.

6. The method according to claim 1, characterized in that, In step (4), it also includes: in the display interface of the computer, according to the calculated accumulation position of the hydrate center and the fitting radius of the hydrate, display the simulation diagram of the predicted position of the hydrate in the specimen in real time.

7. A system for real-time detecting the accumulation state of hydrates in a core sample, characterized in that, It includes a sound velocity detection device, a specimen displacement device, a pulse generator / receiver, a data acquisition card, and a computer; among them, The sound velocity detection device includes a main body component, a transmitting ultrasonic probe, and a receiving ultrasonic probe; the main body component has a horizontally arranged cylindrical specimen channel, the transmitting ultrasonic probe and the receiving ultrasonic probe are relatively installed on the upper and lower surfaces of the main body component, and the connection line of their centers coincides with the maximum diameter direction of the specimen channel; The specimen displacement device is used to pass the core specimen through the specimen channel of the sound velocity detection device at a set speed; A pulse generator / receiver is connected to an ultrasonic transmitting probe and an ultrasonic receiving probe respectively through signal lines, and is used to trigger the ultrasonic transmitting probe to emit pulsed acoustic waves radially along the core sample and receive the acoustic signal after penetrating the core sample; A data acquisition card is connected to the pulse generator / receiver and a computer respectively through signal lines, and is used to acquire the acoustic signal received by the pulse generator / receiver and send it to the computer for processing; The computer has built-in software function modules, which are used to draw a real-time waveform diagram and perform denoising and normalization processing to obtain the input data of the core sample, input it into a trained neural network, calculate the prediction result of the hydrate accumulation state, and display it on the display interface of the software function module; the prediction result includes the accumulation position of the hydrate center in the cross-section of the core column, the fitting radius of the hydrate, and the acoustic velocity of the core.

8. The system according to claim 7, wherein Both the ultrasonic transmitting probe and the ultrasonic receiving probe are piezoelectric crystal longitudinal wave probes, which are used to realize the mutual conversion between mechanical acoustic signals and electrical signals.

9. The system according to claim 7, wherein The main bodies of the ultrasonic transmitting probe and the ultrasonic receiving probe are cylindrical, with signal line interfaces on the side and are connected to the pulse generator / receiver through signal lines; circular bosses are provided on the bottom surface of the probe, and the surface of the bosses is used as the ultrasonic contact surface and is pressed into the circular concave groove on the surface of the main body member; the probe is fixed on the main body member by a probe pressure bridge and fixing studs; the cross-section of the probe pressure bridge is bridge-shaped, and circular through holes are provided at both bridge feet for passing through the fixing studs; a circular through hole is provided in the middle of the probe pressure bridge, and the probe is pressed by a fastening stud and a probe cap.

Images