Method for establishing accurate shot-geophone positioning model

Abstract

The invention discloses a method for establishing an accurate shot-geophone positioning model. The method comprises the following steps: S1, acquiring parameters; s2, representing the relation between the horizontal offset x and the vertical offset y of each detection point according to the obtained parameters and by using a polynomial coefficient; s3, representing a seabed interface by using a polynomial coefficient; s4, acquiring features for network training according to the representation information of the seabed interface; s5, calculating a corresponding label through the features; s6, according to the polynomial coefficient vector, the direct wave travel and the submarine reflected wave travel, network training is carried out on the polynomial coefficient vector, the direct wave travel and the submarine reflected wave travel, and a trained neural network function is obtained; step S7, according to the trained neural network function, reconstructing deep-towed earthquake shot-geophone positioning; the problem of low calculation efficiency of deep dragging earthquake shot-geophone positioning is solved.

Description

Technical Field

[0001] The invention relates to the field of methods for establishing a gun inspection positioning model, and in particular to a method for establishing an accurate gun inspection positioning model. Background Art

[0002] Deep-towed multi-channel seismic detection technology is a multi-channel seismic detection technology that deploys the source and receiving cables at a certain height near the seabed. Because it is deployed at a certain height near the seabed, the lateral and vertical resolution of offshore seismic exploration is greatly improved. Since the 1980s, with the urgent demand for offshore energy exploration and submarine geological disasters, deep-towed seismic technology has developed rapidly. It can provide good data for submarine natural gas hydrate exploration, landslide hazard assessment, submarine tectonic activities and other research. With the development of the instrumentation industry and the advancement of data processing technology, deep-towed multi-channel seismic detection technology will surely play a greater role in future geophysical exploration.

[0003] Accurate shot positioning plays a vital role in marine seismic data processing, which greatly affects the resolution and signal-to-noise ratio of the seismic profile. However, conventional offshore multi-channel seismic detection has acoustic waterfowl, positioning tail markers and other equipment to constrain the relative position of the source and the detector, while deep-towed seismic requires very high shot positioning accuracy. The current deep-towed seismic shot positioning methods mostly use swarm intelligence algorithm optimization, which requires multiple optimizations in multiple shot sets. As the number of excitations increases, the time cost increases significantly.

[0004] Aiming at the problem of low calculation efficiency of deep-towed seismic shot positioning, a method for establishing an accurate shot positioning model is proposed. Conventional calculation methods require calculating the positioning of each shot set one by one, and this method breaks through this limitation.

[0005] Technical solutions of the prior art

[0006] The existing deep-towed seismic shot positioning method converts the shot positioning problem into an optimization problem with each track pitch angle as the optimization target in a plane rectangular coordinate system, based on the fixed relationship between the offset distance between the source and the detection point, the track spacing, and the constraints on the arrival of the direct wave and the seabed reflection wave.

[0007] In order to describe the position of the geophone relative to the earthquake source, a plane rectangular coordinate system is established with the earthquake source as the origin ( Figure 1 ), assuming that the source is connected to the detector 1 by a straight line, the distance is the minimum offset distance L, the detectors are also connected by a straight line, the distance is the track spacing D, and the number of seismic tracks is j. The coordinates of the detector are determined in this way. The coordinates of the detector 1 (x 1 ,y 1 ) is: (Lcosθ 1 ,Lsinθ 1), the horizontal and vertical coordinates of the i-th detector (i≥2) are:

[0008]

[0009] Among them, θ 1 is the angle between the line connecting the source and the detector 1 and the horizontal direction, θ i is the angle between the line connecting the i-th detector and the i-1-th detector and the horizontal direction, and the vector θ = [θ 1 ,θ 2 ,...,θ j-1 ,θ j ] T is the pitch angle vector of each channel, each θ corresponds to a tow cable attitude (gun inspection positioning)

[0010] The arrival time of the direct wave and the seabed reflection wave reflects the relationship between the seabed interface morphology and the gun inspection positioning, and therefore serves as a constraint condition for this problem. Since the seabed interface depth is known, different pitch angles correspond to different arrival times of the direct wave and the seabed reflection wave. The root mean square error (RMSE) of the theoretical and actual arrival times of the direct wave and the seabed reflection wave is used as the objective function of the optimization algorithm to calculate a more accurate gun inspection positioning.

[0011] Disadvantages of existing technology

[0012] First, according to Figure 1 As shown in the figure, this gun inspection system model assumes that the traces are straight lines, while the actual situation should be curved lines. This is determined by the inherent properties of the receiving cable, and will lead to cumulative positioning errors as the number of traces increases.

[0013] Secondly, due to the movement and extension and retraction of the deep-drag source and the detector array, the shot detection positioning is different each time the seismic wave is excited. Therefore, the shot detection positioning needs to be calculated once for each excitation, which greatly increases the calculation time cost. Summary of the invention

[0014] In order to solve the problems existing in the prior art, the present invention provides a method for establishing an accurate shot detection positioning model, which solves the problem of low calculation efficiency of deep-towed seismic shot detection positioning.

[0015] The specific technical solutions are as follows:

[0016] A method for establishing an accurate gun inspection positioning model comprises the following steps:

[0017] Step S1: Obtain parameters;

[0018] Step S2: Characterize the relationship between the horizontal offset x and the vertical offset y of each detection point according to the acquired parameters and using the polynomial coefficients;

[0019] Step S3: characterizing the seafloor interface using polynomial coefficients;

[0020] Step S4: acquiring features for network training based on the representation information of the seabed interface;

[0021] Step S5: Calculate corresponding labels through features;

[0022] Step S6: training a network according to the polynomial coefficient vector, the direct wave travel, and the seabed reflected wave travel to obtain a trained neural network function;

[0023] Step S7: Reconstruct the deep-towed seismic shot detection positioning according to the trained neural network function.

[0024] Preferably, the parameters include the number of source excitations N, the depth measurement value H from the sea level at the time of source excitation D The measured value of the height of the earthquake source from the seafloor interface H, and the observed value of the depth of the seafloor interface from the sea level at the time of earthquake source excitation H D +H, the travel time observation value T of the direct wave and the seabed reflection wave recorded at each detection point D and T F , the speed of sound v of water near the bottom.

[0025] Preferably, step S2 includes the following sub-steps:

[0026] Sub-step S21: Obtain the polynomial matrix P(x), which is expressed as:

[0027] P(x)=R T X

[0028] Where: R = [r n r n-1 ...r 2 r 1 ] T is the polynomial coefficient vector, X = [x n x n-1 ...x 2 x] T is the power vector of the variable;

[0029] Sub-step S22: The derivative of the polynomial is P'(x). According to the arc length formula, the coordinates of detector 1 can be expressed as (x 1 ,P(x 1 )), where x 1 satisfy:

[0030]

[0031] The coordinates of the i-th detector (i ≥ 2) are (x i ,P(x i)), where each x i satisfy:

[0032]

[0033] Therefore, the coefficient vector R is used to represent the horizontal offset x and the vertical offset y of each detection point.

[0034] Preferably, step S3 specifically includes:

[0035] The seafloor interface depth is fitted by a k-order polynomial F(x), which can be expressed as

[0036] F(x)=x k x k +s k-1 x k-1 +...+s 2 x 2 +s 1 x+s 0

[0037] The matrix representation of the polynomial F(x) is:

[0038] F(x)=S T X

[0039] Where: S = [s k s k-1 ...s 1 s 0 ] T is the polynomial coefficient vector, X = [x k x k-1 ... x1] T is a power vector of variables. The number of polynomial coefficients of the seabed interface depth is k+1. The coefficient vector S is used to represent the seabed interface depth.

[0040] Preferably, step S4 includes the following sub-steps:

[0041] Sub-step S41: setting a reasonable range of the earthquake source height from the bottom;

[0042] Sub-step S42: setting a reasonable range of vertical offset between the earthquake source and the farthest offset;

[0043] Sub-step S43: setting a reasonable range of seabed slope angle;

[0044] Sub-step S44: obtaining multiple groups of polynomial coefficients R and coefficients S that meet the conditions by means of target optimization.

[0045] Preferably, step S5 includes the following sub-steps:

[0046] Sub-step S51: Calculate the direct wave travel time T according to the offset x and offset y represented by multiple sets of polynomial coefficients R D =[d 1 d 2 ...d j-1 d j ] T ;

[0047] Sub-step S52: Calculate the travel time T of the seabed reflection wave using the Fermat principle based on multiple sets of polynomial coefficients R and S and the acoustic velocity v near the seabed F =[f 1 f 2 ...f j-1 f j ] T .

[0048] Preferably, step S6 specifically includes:

[0049] The feature is: Ι = [R, S], the label is: Ο = [T D ,T F ], the mapping relationship is:

[0050] O=f θ (I)

[0051] Among them, f θ is the trained neural network function.

[0052] Preferably, step S7 includes the following sub-steps:

[0053] Sub-step S71: Characterize the H at N excitation moments according to step S3 D +H polynomial and extract the polynomial coefficient S 0 ;

[0054] Sub-step S72: Combine the T obtained in step S1 D and T F , input the network to calculate the corresponding coefficient R, and further combine step S2 and step S5 to obtain the gun inspection positioning.

[0055] The beneficial effects of the method for establishing an accurate gun inspection positioning model of the present invention are as follows:

[0056] 1. The present invention uses polynomial coefficients to characterize the relative positions of the earthquake source and the detection points in deep-drag earthquakes. Based on the parametric expression of the polynomial, it can effectively solve the problems of large cumulative errors and low calculation efficiency in the existing methods.

[0057] 2. The present invention improves computational efficiency: the traditional method requires that shot inspection positioning be performed one by one each time a seismic wave is excited. However, the present invention uses a neural network to train a large number of polynomial coefficients, and can directly derive accurate shot inspection positioning after obtaining the arrival time of direct waves and seabed reflection waves, thereby greatly reducing the computational time.

[0058] 3. The present invention reduces cumulative errors: The prior art assumes that the lines between the detectors are straight lines, ignoring the influence of the arc of the tow cable posture. The present invention uses polynomials to fit the spatial position of the detector array, which is more in line with the actual situation and reduces the cumulative positioning error.

[0059] 4. The model of the present invention has strong scalability: By introducing polynomial coefficients to characterize the gun detection positioning, the method of the present invention has good scalability. In different situations, it only needs to adjust the order of the polynomial or the range of training data to adapt to different source excitation depths, detection point distributions and seabed interface morphologies.

[0060] 5. The present invention is not strongly dependent on the underwater acoustic velocity near the seabed: the existing technology has very high requirements for the accuracy of the underwater acoustic velocity. When there is a slight difference between the set sound velocity and the actual sound velocity, it can still affect the overall position of multiple detection points. In the present invention, since the underwater acoustic velocity is not directly reflected in the output and input ends of the neural network, it has better fault tolerance for the underwater acoustic velocity. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic diagram of the conventional gun inspection positioning description method of the present invention.

[0062] Figure 2 It is a schematic diagram of the method for describing the gun inspection positioning of the present invention.

[0063] Figure 3 It is the technical roadmap of the solution of the present invention.

[0064] Figure 4 It is a schematic diagram of the propagation of seabed reflection waves in a single sample of the present invention.

[0065] Figure 5 It is a relative position diagram of the source and receiver points of multiple samples of the present invention.

[0066] Figure 6 This is a positioning error distribution diagram predicted by the network of the present invention. DETAILED DESCRIPTION

[0067] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0068] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0069] A method for establishing an accurate gun inspection positioning model comprises the following steps:

[0070] Step S1: Obtain parameters;

[0071] Step S2: Characterize the relationship between the horizontal offset x and the vertical offset y of each detection point according to the acquired parameters and using the polynomial coefficients;

[0072] Step S3: characterizing the seafloor interface using polynomial coefficients;

[0073] Step S4: acquiring features for network training based on the representation information of the seabed interface;

[0074] Step S5: Calculate corresponding labels through features;

[0075] Step S6: training a network according to the polynomial coefficient vector, the direct wave travel, and the seabed reflected wave travel to obtain a trained neural network function;

[0076] Step S7: Reconstruct the deep-towed seismic shot detection positioning according to the trained neural network function.

[0077] The parameters of this implementation scheme include the number of source excitations N, the depth measurement value H from the sea level at the time of source excitation D The measured value of the height of the earthquake source from the seafloor interface H, and the observed value of the depth of the seafloor interface from the sea level at the time of earthquake source excitation H D +H, the travel time observation value T of the direct wave and the seabed reflection wave recorded at each detection point D and T F , the speed of sound v of water near the bottom.

[0078] Step S2 of this embodiment includes the following sub-steps:

[0079] Sub-step S21: Obtain the polynomial matrix P(x), which is expressed as:

[0080] P(x)=R T X

[0081] Where: R = [r n r n-1 ...r 2 r 1 ]T is the polynomial coefficient vector, X = [x n x n-1 ...x 2 x] T is the power vector of the variable;

[0082] Sub-step S22: The derivative of the polynomial is P'(x). According to the arc length formula, the coordinates of detector 1 can be expressed as (x 1 ,P(x 1 )), where x 1 satisfy:

[0083]

[0084] The coordinates of the i-th detector (i ≥ 2) are (x i ,P(x i )), where each x i satisfy:

[0085]

[0086] Therefore, the coefficient vector R is used to represent the horizontal offset x and the vertical offset y of each detection point.

[0087] Step S3 of this embodiment is specifically as follows:

[0088] The seafloor interface depth is fitted by a k-order polynomial F(x), which can be expressed as

[0089] F(x)=s k x k +s k-1 x k-1 +...+s 2 x 2 +s 1 x+s 0

[0090] The matrix representation of the polynomial F(x) is:

[0091] F(x)=S T X

[0092] Where: s = [s k s k-1 ...s 1 s 0 ] T is the polynomial coefficient vector, X = [x k x k-1 ... x1] T is a power vector of variables. The number of polynomial coefficients of the seabed interface depth is k+1. The coefficient vector S is used to represent the seabed interface depth.

[0093] Step S4 of this embodiment includes the following sub-steps:

[0094] Sub-step S41: setting a reasonable range of the earthquake source height from the bottom;

[0095] Sub-step S42: setting a reasonable range of vertical offset between the earthquake source and the farthest offset;

[0096] Sub-step S43: setting a reasonable range of seabed slope angle;

[0097] Sub-step S44: obtaining multiple groups of polynomial coefficients R and coefficients S that meet the conditions by means of target optimization.

[0098] Step S5 of this embodiment includes the following sub-steps:

[0099] Sub-step S51: Calculate the direct wave travel time T according to the offset x and offset y represented by multiple sets of polynomial coefficients R D =[d 1 d 2 ...d j-1 d j ] T ;

[0100] Sub-step S52: Calculate the travel time T of the seabed reflection wave using the Fermat principle based on multiple sets of polynomial coefficients R and S and the acoustic velocity v near the seabed F =[f 1 f 2 ...f j-1 f j ] T .

[0101] Step S6 of this embodiment is specifically as follows:

[0102] The feature is: Ι = [R, S], the label is: Ο = [T D ,T F ], the mapping relationship is:

[0103] O=f θ (I)

[0104] Among them, f θ is the trained neural network function.

[0105] Step S7 of this embodiment includes the following sub-steps:

[0106] Sub-step S71: Characterize the H at N excitation moments according to step S3 D +H polynomial and extract the polynomial coefficient S 0 ;

[0107] Sub-step S72: Combine the T obtained in step S1 D and T F , input the network to calculate the corresponding coefficient R, and further combine step S2 and step S5 to obtain the gun inspection positioning.

[0108] When this implementation plan is implemented,

[0109] (1) Obtain the number of earthquake source excitations N; obtain the depth measurement value H from the sea level at the time of earthquake source excitation D and the measured value H of the height of the earthquake source from the seafloor interface; obtain the observed value H of the depth of the seafloor interface from the sea level at the time of earthquake source excitation D +H; obtain the travel time observation value T of the direct wave and the seabed reflection wave recorded at each detection point D and T F ; Get the near-bottom water sound velocity v;

[0110] (2) The polynomial coefficients are used to characterize the relationship between the horizontal offset x and the vertical offset y of each detection point.

[0111] like Figure 2 As shown, the source and multiple detectors can be reflected in an n-order polynomial P(x). Since the polynomial passes through the origin of the coordinate system, it can be expressed as:

[0112] P(x)=r n x n +r n-1 x n-1 +...+r 2 x 2 +r 1 x (2)

[0113] The matrix representation of the polynomial P(x) is:

[0114] P(x)=R T X (3)

[0115] Where: R = [r n r n-1 ... r 2 r 1 ] T is the polynomial coefficient vector, X = [x n x n-1 ... x 2 x] T is the power vector of the variable. It can be seen that the number of polynomial coefficients in this model is n.

[0116] The derivative of this polynomial is P'(x). According to the arc length formula, the coordinates of detector 1 can be expressed as (x 1 ,P(x 1)), where x 1 satisfy:

[0117]

[0118] Similarly, the coordinates of the i-th detector (i≥2) are (x i ,P(x i )), where each x i satisfy:

[0119]

[0120] Therefore, the coefficient vector R is used to represent the horizontal offset x and the vertical offset y of each detection point.

[0121] (3) Using the polynomial coefficients to characterize the seafloor interface

[0122] Assume that the seafloor interface depth is fitted by a k-order polynomial F(x), which can be expressed as:

[0123] F(x)=s k x k +s k-1 x k-1 +...+s 2 x 2 +s 1 x+s 0 (6)

[0124] The matrix representation of the polynomial F(x) is:

[0125] F(x)=S T X (7)

[0126] Where: S = [s k s k-1 ...s 1 s 0 ] T is the polynomial coefficient vector, X = [x k x k-1 ... x 1] T is a power vector of variables. It can be seen that the number of polynomial coefficients of the seafloor interface depth is k+1.

[0127] The coefficient vector S is thus used to characterize the depth of the seafloor interface.

[0128] (4) Obtaining features for network training

[0129] Set a reasonable range of the height of the earthquake source from the bottom; set a reasonable range of the vertical offset between the earthquake source and the farthest offset; set a reasonable range of the seabed slope angle; obtain multiple sets of polynomial coefficients R and coefficients S that meet the conditions through target optimization.

[0130] (5) Calculate the corresponding label through features

[0131] Based on the offset x and offset y represented by multiple sets of polynomial coefficients R, the direct wave travel time T is calculated. D =[d 1 d 2 ...d j-1 d j ] T ; Based on multiple sets of polynomial coefficients R and S, combined with the near-seabed water sound velocity v, the Fermat principle is used to calculate the travel time T of the seabed reflection wave F =[f 1 f 2 ... f j-1 f j ] T .

[0132] (6) Training the network

[0133] The feature is: I = [R, S], the label is: O = [T D ,T F ], the mapping relationship is:

[0134] O=f θ (I) (8)

[0135] Among them, f θ is the trained neural network function.

[0136] (7) Deep-towed seismic shot positioning reconstruction

[0137] According to step S3, H representing N excitation moments D +H polynomial and extract the polynomial coefficient S 0 , combined with T obtained in step S1 D and T F , input the network to calculate the corresponding coefficient R, and further combine step S2 and step S5 to obtain the gun inspection positioning.

[0138] In this example, the deep-towed seismic system is equipped with an altimeter and a depth gauge on the towed body, which continuously records the height of the towed body from the bottom and the sinking depth during operation. The system has 48 receiving cables with a channel spacing of 3.125m and a minimum offset of 12.5m.

[0139] Figure 3 This is the technical roadmap of the solution of the present invention. Combined with the figure, it can be seen that deep-towed seismic shot detection positioning based on neural network consists of two parts, namely network training and prediction using the network.

[0140] Before network training, we first need to construct a simulation data set (taking the fourth-order polynomial fitting of the earthquake source and the receiver point, and the third-order polynomial fitting of the seafloor interface height observation value as an example). A single sample of the data set contains ( Figure 4 ):

[0141] (1) 4 polynomial coefficients for fitting the source and receiver points;

[0142] (2) 4 polynomial coefficients for fitting the seafloor interface height;

[0143] (3) The travel times of the 48 direct waves corresponding to this sample;

[0144] (4) The travel time of the 48 seafloor reflection waves corresponding to this sample.

[0145] Combined with the relative positions of the source and the receiver points and the depth variation range of the seafloor interface during actual operations, the range of the polynomial coefficients is further constrained to construct a more suitable data set. For example, the height difference between the source height and the farthest offset receiver point should not be too large, and the elevation angle of a single receiver point should not be too large.

[0146] Figure 5 The difference between the source height and the maximum offset detection point height is limited to within ±50m, and the pitch angle of a single detection point is limited to within ±30°. Similarly, the depth change of the seafloor interface should not be too large, and similar constraints are adopted.

[0147] 100,000 samples were generated in the above manner, and the 48 direct wave travel times corresponding to the polynomial coefficients fitting the source and receiver points and the polynomial coefficients fitting the seafloor interface height in a single sample were calculated. The 48 seafloor reflection wave travel times were calculated using the Fermat principle.

[0148] 80% of the dataset was used as the training set, 10% as the test set, and 10% as the validation set. A single-layer feedforward neural network with 20 hidden layer neurons was used as the input. The mean square error was used as the loss function, and 100 training rounds were set. After training, the performance of the neural network model on the validation set achieved satisfactory accuracy, with a mean square error (MSE) of 3.03×10 -7 .

[0149] Using the trained network, the detector location is predicted when the seabed interface, direct wave and seabed reflection wave travel time are known. The positioning error is expressed as the difference between the predicted value and the theoretical value of the detection point location. 100 new samples are generated for prediction. The results show that the average positioning error of each channel is less than 0.162m, and the error distribution is as follows: Figure 6(a) When there is an error in picking up the direct wave or the seabed reflected wave (assuming there is a Gaussian distribution with an expected value of 2ms and a variance of 1), the overall positioning error is less than 0.3m, and the error distribution is as follows: Figure 6 (b).

[0150] The key technical points and points to be protected of the present invention are the method of using polynomials to describe artillery detection positioning and the artillery detection positioning calculation method using polynomial coefficients as neural network input and output.

Claims

1. A method for establishing an accurate gun inspection positioning model, characterized in that: The following steps are involved: Step S1: Obtain parameters; Step S2: Characterize the relationship between the horizontal offset x and the vertical offset y of each detection point according to the acquired parameters and using the polynomial coefficients; Step S3: characterizing the seafloor interface using polynomial coefficients; Step S4: acquiring features for network training based on the representation information of the seabed interface; Step S5: Calculate corresponding labels through features; Step S6: training a network according to the polynomial coefficient vector, the direct wave travel, and the seabed reflected wave travel to obtain a trained neural network function; Step S7: Reconstruct the deep-towed seismic shot detection positioning according to the trained neural network function.

2. The method for establishing an accurate gun inspection positioning model according to claim 1, characterized in that: The parameters include the number of source excitations N, the depth measurement value H from the sea level at the time of source excitation D The measured value of the height of the earthquake source from the seafloor interface H, and the observed value of the depth of the seafloor interface from the sea level at the time of earthquake source excitation H D +H, the travel time observation value T of the direct wave and the seabed reflection wave recorded at each detection point D and T F , the speed of sound v of water near the bottom.

3. The method for establishing an accurate gun inspection positioning model according to claim 1, characterized in that: The step S2 comprises the following sub-steps: Sub-step S21: Obtain the polynomial matrix P(x), which is expressed as: P(x)=R T X Where R = [r n r n-1 ...r2 r1] T is the polynomial coefficient vector, X = [x n x n-1 ...x 2 x] T is the power vector of the variable; Sub-step S22: The derivative of the polynomial is P'(x). According to the arc length formula, the coordinates of detector 1 can be expressed as (x1, P(x1)), where x1 satisfies: The coordinates of the i-th detector (i ≥ 2) are (x i ,P(x i )), where each x i satisfy: The coefficient vector R is used to represent the horizontal offset x and vertical offset y of each detection point.

4. The method for establishing an accurate gun inspection positioning model according to claim 1, characterized in that: The step S3 is specifically as follows: The seafloor interface depth is fitted by a k-order polynomial F(x), which can be expressed as F(x) = s k x k +s k-1 x k-1 +...+s2x 2 +s1x+s0 The matrix representation of the polynomial F(x) is: F(x)=S T X Where: S = [s k s k-1 ... s1 s0] T is the polynomial coefficient vector, X = [x k x k-1 ... x1] T is a power vector of variables. The number of polynomial coefficients of the seabed interface depth is k+1. The coefficient vector S is used to represent the seabed interface depth.

5. The method for establishing an accurate gun inspection positioning model according to claim 1, characterized in that: The step S4 comprises the following sub-steps: Sub-step S41: setting a reasonable range of the earthquake source height from the bottom; Sub-step S42: setting a reasonable range of vertical offset between the earthquake source and the farthest offset; Sub-step S43: setting a reasonable range of seabed slope angle; Sub-step S44: obtaining multiple groups of polynomial coefficients R and coefficients S that meet the conditions by means of target optimization.

6. The method for establishing an accurate gun inspection positioning model according to claim 1, characterized in that: The step S5 comprises the following sub-steps: Sub-step S51: Calculate the direct wave travel time T according to the offset x and offset y represented by multiple sets of polynomial coefficients R D =[d1 d2...d j-1 d j ] T ; Sub-step S52: Calculate the travel time T of the seabed reflection wave using the Fermat principle based on multiple sets of polynomial coefficients R and S and the acoustic velocity v near the seabed F =[f1 f2...f j-1 f j ] T .

7. The method for establishing an accurate gun inspection positioning model according to claim 1, characterized in that: The step S6 is specifically as follows: The feature is: Ι = [R, S], the label is: Ο = [T D ,T F ], the mapping relationship is: O=f θ (I) Among them, f θ is the trained neural network function.

8. The method for establishing an accurate gun inspection positioning model according to claim 1, characterized in that: The step S7 comprises the following sub-steps: Sub-step S71: Characterize the H at N excitation moments according to step S3 D +H's polynomial and extract the polynomial coefficient S0; Sub-step S72: Combine the T obtained in step S1 D and T F , input the network to calculate the corresponding coefficient R, and further combine step S2 and step S5 to obtain the gun inspection positioning.

Images