Gravimeter vibration compensation method jointly using iterative computation and BP neural network

Through the combined method of iterative calculation and BP neural network, the ADMM algorithm is used to optimize the vibration compensation model and delay gain model, which solves the problem of insufficient vibration compensation accuracy of the gravity meter in complex vibration environments, and achieves high-precision gravity measurement.

CN119001885APending Publication Date: 2024-11-22ROCKET FORCE UNIV OF ENG
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410926769.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-11
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The vibration compensation accuracy of the prior art gravitational instruments is insufficient in complex vibration environments, and the model simplification cannot accurately describe the real vibration transmission process, resulting in poor compensation accuracy.

Method used

The combined method of iterative calculation and BP neural network is adopted to establish a vibration compensation optimization model through the ADMM algorithm to obtain the optimal delay coefficient and gain coefficient, and the remaining error is recompensated by the BP neural network to improve the accuracy of gravity measurement.

Benefits of technology

It significantly improves the accuracy of gravity measurement in vibration environments, avoids the problems of insufficient compensation or overcorrection, and enhances the adaptability of the gravity instrument in complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119001885A_ABST
    Figure CN119001885A_ABST
Patent Text Reader

Abstract

The invention discloses a gravimeter vibration compensation method jointly using iterative computation and a BP neural network, and the method comprises the steps: simplifying a vibration transfer function into a time delay gain model based on the influence mechanism research of a vibration environment on a gravimeter, obtaining a global optimal model coefficient through employing a neighborhood term improved alternating direction multiplier method (ADMM), and carrying out the calculation of the model coefficient. And carrying out iterative compensation on the gravity measurement value. And constructing a data set of a BP neural network according to the gravity output and the vibration detection signal after iteration compensation and the standard output of the gravimeter in a vibration-free environment to carry out vibration compensation model training, wherein the obtained model is used for compensating the residual error again to obtain a final vibration compensation result. The vibration compensation algorithm combining iterative computation and the BP neural network can effectively isolate the influence of the vibration environment on gravity measurement, and the environmental adaptability of the spring type relative gravimeter is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of gravimeter vibration compensation, and in particular relates to a gravimeter vibration compensation method using iterative calculation and BP neural network in combination. Background Art

[0002] High-precision field measurement of gravity information is like doing a CT scan of the earth to see through the underground structure. Accurate gravity distribution information has become a national strategic resource and has extremely important application value in the fields of bridge and tunnel construction, earthquake prediction, resource exploration, and weapon launch.

[0003] The zero-length spring relative gravimeter, represented by the domestically produced CHZ-Ⅱ gravimeter, has the advantages of small zero bias, high accuracy, and strong reliability. Its static measurement accuracy has reached the international advanced level. However, under complex and noisy field conditions, the vibration environment greatly affects its measurement accuracy and reliability. The two springs and six wires in the gravimeter effectively limit the movement of the mass body except the vertical translation, and the axisymmetric structure makes the resultant force of the horizontal acceleration on the tubular mass block zero to suppress the horizontal cross-coupling effect. Therefore, this type of gravimeter is insensitive to horizontal vibration, and the vertical vibration acceleration is the main bottleneck limiting the accuracy of gravity measurement.

[0004] Vibration isolation and vibration compensation are two main ways to suppress the acceleration of vibration environment disturbance. Vibration isolation has higher accuracy on a stable platform, but its structure is bulky, expensive, difficult to implement, and has poor anti-interference ability, and is not suitable for use in complex vibration environments. Vibration compensation has the characteristics of simple structure, portability, flexibility, and low cost. It is widely used and rapidly developed in highly dynamic and complex measurement environments. Vibration compensation applied to airborne terahertz SAR imaging effectively suppresses the phase error introduced by high-frequency vibration and improves the imaging quality. The absolute ranging technology based on frequency scanning interferometry has a significant improvement in ranging accuracy in natural environments after vibration compensation. Vibration compensation simplifies the vibration transmission process into a time delay gain model during the measurement of the absolute gravimeter and searches the coefficients of the model through optimization algorithms such as sparrow search algorithm, two-dimensional golden section algorithm, correlation analysis, PSO, etc., which significantly improves the accuracy of gravity measurement in complex vibration environments.

[0005] In the field of relative gravity measurement, the research on vibration compensation has not attracted widespread attention. At present, the research on vibration compensation methods uses a single calculation to obtain the global optimal solution. This process inevitably leads to under-compensation or over-compensation, resulting in poor compensation accuracy. In addition, the simplification of the model cannot accurately describe the actual vibration transmission process, and non-modeled errors are often ignored. Summary of the invention

[0006] The purpose of the present invention is to provide a gravimeter vibration compensation method which combines iterative calculation and BP neural network, has high compensation accuracy and can fully suppress the influence of vibration environment on gravity measurement.

[0007] The technical solution adopted by the present invention is to use a gravity meter vibration compensation method combining iterative calculation and BP neural network, which is implemented according to the following steps:

[0008] Step 1, obtaining preliminary gravity measurement values ​​of a zero-length spring relative gravimeter and a flexible accelerometer gravimeter and detection values ​​of a vibration environment, and filtering the detection values ​​of the vibration environment to obtain vibration acceleration;

[0009] Step 2: Establish a vibration compensation optimization model through vibration transfer relationship and ADMM algorithm;

[0010] Step 3: Obtain the optimal delay coefficient and gain coefficient in the optimization model through the ADMM algorithm improved by the neighboring term;

[0011] Step 4: Compensate the initial gravity measurement value and update the detection value of the vibration environment. Through iterative calculation, the optimal compensated gravity measurement value and the iteratively updated vibration environment detection value are obtained, and the training sample is constructed in combination with the standard data of the gravimeter in a vibration-free environment.

[0012] Step 5: Construct a BP neural network, and train the BP neural network model with training samples. The obtained model is used to compensate the residual errors outside the vibration compensation optimization model again to obtain the final vibration compensation result of the gravimeter.

[0013] The present invention is also characterized in that:

[0014] The process of obtaining the preliminary gravity measurement values ​​of the zero-length spring relative gravimeter and the flexible accelerometer gravimeter and the detection values ​​of the vibration environment in step 1 is: fix the zero-length spring relative gravimeter and the flexible accelerometer gravimeter and perform static measurement to obtain the preliminary gravity measurement values ​​and the detection values ​​of the vibration environment.

[0015] The specific filtering of the vibration environment detection value in step 1 is as follows: a low-pass filter with a cutoff frequency of 0.05 Hz is designed to filter the vibration environment detection value to obtain the vibration acceleration data a in the same frequency band as the initial gravity measurement value. f .

[0016] The specific process of step 2 is as follows:

[0017] Step 2.1: Use the delay gain model to simplify the real transfer function. The gain coefficient and delay coefficient remain unchanged during the current measurement process. Then:

[0018] a c = k·af (t+τ);

[0019] where a c is the vibration compensation signal used to approximate the actual vibration condition of the mass block, t is the time mark of the disturbance acceleration measurement signal, k is the gain coefficient, and τ is the delay coefficient;

[0020] Step 2.2: According to the basic principle of ADMM algorithm, the basic structure and augmented Lagrangian function of the vibration compensation optimization model constructed based on the delay gain model are:

[0021]

[0022]

[0023]

[0024] f(k)=std(g m -a c (k));

[0025] where g m is the measured value of gravity, g r is the true value of gravity, λ is the column vector of multiplier variables; σ is a constant penalty factor, ||·||2 is the vector two-norm operator, h(τ) and f(k) are the objective functions of the delay coefficient τ and the gain coefficient k respectively, corr is the correlation calculation function, and std is the standard deviation calculation function.

[0026] The specific process of step 3 is as follows:

[0027] Step 3.1, initialize the parameters in the ADMM algorithm: constant penalty factor σ∈(0,1), vector update coefficient β∈[0.5,1), initial multiplier variable λ0 is a column vector with all elements set to 1, golden ratio Γ and K are the feasible domains of τ and k respectively, and τ0∈Γ, k0∈K, the neighboring term coefficients S and T are set to random numbers in (0, 0.5), and the maximum number of iterations of the algorithm is set;

[0028] Step 3.2: Update the parameters of the τ-subproblem based on the improvement of the golden ratio neighboring term. At this moment, only the τ-subproblem is in the calculation state. After the update of this problem is completed, it will be shared with the k-subproblem. The update formula is:

[0029]

[0030]

[0031] Where the subscript n is the update algebra, u represents the process variable τ is golden sectioned and u0=τ0,η nis the update weight of the neighboring items of τ, y represents the variable to be calculated;

[0032] Step 3.3: Half update of the column vector of multiplier variables; the update formula is:

[0033]

[0034] Step 3.4: Update the parameters of the k-subproblem according to the algorithm principle. After the calculation is completed, share it with another subproblem for the next round of optimization calculation until the maximum number of iterations is reached. The parameter update process is as follows:

[0035]

[0036] where μ n is the update weight of k neighboring items;

[0037] Step 3.5: Update the calculation of the multiplier variables and other process variables. After reaching the maximum number of updates, the global optimal model coefficient gain coefficient k is obtained. opt and the delay factor τ opt ; The update formula is:

[0038]

[0039]

[0040] ξ n+1 =ρμ n+1 .

[0041] In step 4, the initial gravity measurement value is compensated, and the detection value of the vibration environment is updated, that is, the initial gravity measurement value and the detection value of the vibration environment are iteratively calculated by optimizing the optimal delay coefficient and gain coefficient in the model. The specific process is:

[0042] a c-opt =k opt ·a f (t+τ opt );

[0043] g m =g m -a c-opt ;

[0044] a f =a f -a c-opt .

[0045] The iteration termination condition in step 4 is:

[0046] The difference in accuracy between two consecutive updated preliminary gravity measurements is less than 1μGal or reaches the maximum number of iterations of the algorithm.

[0047] In step 5, the input layer nodes of the BP neural network are 2, the output layer is 1 node, a single hidden layer network structure is selected and the number of nodes is set to 6.

[0048] The beneficial effects of the present invention are:

[0049] 1. The present invention fuses the data of two different types of gravimeters and proposes a vibration compensation method for a zero-length spring relative gravimeter based on a simplified delay gain model, which has high compensation accuracy.

[0050] 2. The present invention improves the optimization effect and convergence speed by using the improved ADMM algorithm with neighboring terms, and proposes to perform vibration compensation by iterative calculation, which effectively avoids the problem of insufficient or overcorrection of vibration compensation in a single calculation and improves the accuracy of vibration compensation.

[0051] 3. The present invention combines the BP neural network to construct a compensation model for non-modeled errors to compensate the residual errors after the initial compensation, which fully overcomes the disadvantage that gravity measurement is significantly affected by the vibration environment, greatly improves the accuracy of gravity measurement in a vibration environment, and can provide technical support for improving the environmental adaptability of high-precision gravity measurement in the field. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of a gravity meter vibration compensation method using iterative calculation and BP neural network in combination with the present invention;

[0053] Figure 2 This is a mechanical model diagram of a zero-length spring gravity-sensitive unit;

[0054] Figure 3 Provides experimental results of embodiments of the present invention;

[0055] Figure 4 An iterative compensation accuracy evolution diagram of an embodiment of the present invention is provided;

[0056] Figure 5 The present invention provides a vibration compensation experimental result diagram of an embodiment. DETAILED DESCRIPTION

[0057] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Example 1

[0059] The present invention combines the gravimeter vibration compensation method using iterative calculation and BP neural network, such as Figure 1 As shown, follow these steps to implement:

[0060] Step 1: Fix the zero-length spring relative gravimeter and the flexible accelerometer gravimeter and perform static measurement to obtain the preliminary gravity measurement value and the detection value of the vibration environment. The detection value of the vibration environment is filtered as follows: a low-pass filter with a cutoff frequency of 0.05 Hz is designed to filter the detection value of the vibration environment to obtain the vibration acceleration data a in the same frequency band as the preliminary gravity measurement value. f .

[0061] Step 2: The mechanical model of the zero-length spring gravity sensitive unit is as follows: Figure 2 As shown in the figure, the influence mechanism of vibration on gravimeter measurement is studied, the qualitative and quantitative relationship between disturbance acceleration and gravimeter output is revealed, and a theoretical basis is provided for constructing a vibration compensation model.

[0062] The basic principle of gravity measurement satisfies Newton's second law, that is, the gravity on the mass block is equal to the spring force. When gravity changes, the mass block deviates from the original equilibrium position and produces a displacement x. The high-precision capacitive displacement sensor accurately detects x to achieve relative measurement of gravity. According to the principle of moment balance, if the tension of each wire is T, the tension on the two springs is 3T. Figure 2 As shown, after the mass block is displaced x, the wire drawing and the spring will produce a small angle rotation of magnitude α. The displacement x is generally a small amount, and the force balance equation satisfies:

[0063]

[0064] Where k0 is the elastic coefficient of the zero-length spring, L is the drawing length, M is the mass of the mass block, Δg is the change in gravity, and k e is the equivalent coefficient of the elastic sensitive unit. The gravimeter uses a strong damping environment formed by silicone oil and electromagnetic force balance feedback to improve the stability of the mass block movement and can initially suppress the disturbance acceleration. When considering the system damping structure, the motion equation of the mass block is:

[0065]

[0066] where a r is the vertical disturbance acceleration actually received by the mass block, and denote the first and second derivatives of the displacement x, respectively. and They represent the damping force and inertia force on the mass block respectively. The above formula can be transformed into:

[0067]

[0068] in Represents the self-vibration circular frequency of the sensitive component. Use a sinusoidal signal to simulate the vertical disturbance acceleration, then ar The specific form can be expressed as:

[0069]

[0070] in ω z and δ represent the amplitude, frequency and phase of the vertical disturbance acceleration respectively. Then the steady-state solution of displacement x is:

[0071]

[0072] In the formula Under the strong damping effect of silicone oil damping and electromagnetic force compensation, the damping coefficient λ is very large, and the above formula can be simplified to:

[0073]

[0074] It can be seen from this that the gravimeter is not only sensitive to changes in gravitational acceleration, but the vertical disturbance acceleration introduced by vibration will also be directly reflected in the output of the gravimeter. Therefore, using certain means to separate the gravitational acceleration from the carrier disturbance acceleration is a key step in improving the measurement accuracy of the gravimeter in complex and noisy environments.

[0075] The specific process is:

[0076] Step 2.1: The vibration signal after filtering preprocessing cannot represent the actual vibration condition of the zero-length spring gravimeter mass block. There is a complex transfer function relationship between the two. The real transfer function is simplified using the delay gain model. The gain coefficient and delay coefficient remain unchanged during the current measurement process, and then:

[0077] a c = k·a f (t+τ)

[0078] where a c is the vibration compensation signal used to approximate the actual vibration condition of the mass block, t is the time mark of the disturbance acceleration measurement signal, k is the gain coefficient, and τ is the delay coefficient;

[0079] Step 2.2: According to the basic principle of ADMM algorithm, the basic structure and augmented Lagrangian function of the vibration compensation optimization model constructed based on the delay gain model are:

[0080]

[0081]

[0082]

[0083] f(k)=std(g m -ac (k));

[0084] where g m is the measured value of gravity, g r is the true value of gravity, λ is the column vector of multiplier variables; σ is a constant penalty factor, ||·||2 is the vector two-norm operator, h(τ) and f(k) are the objective functions of the delay coefficient τ and the gain coefficient k respectively, corr is the correlation calculation function, and std is the standard deviation calculation function.

[0085] Step 3: Obtain the optimal delay coefficient and gain coefficient in the optimization model through the ADMM algorithm improved by the neighboring term;

[0086] Step 4: Iteratively calculate the initial gravity measurement value and the vibration environment detection value by optimizing the best delay coefficient and gain coefficient in the model. The specific process is as follows:

[0087] a c-opt =k opt ·a f (t+τ opt );

[0088] g m =g m -a c-opt ;

[0089] a f =a f -a c-opt .

[0090] Call the ADMM algorithm again and repeat steps 2.2 to 3.5 for iterative calculation until the difference in gravity signal accuracy between two consecutive compensations is less than 1μGal. After vibration compensation, output the gravity value g at this time. m and vibration detection signal a f , which is the optimally compensated gravity measurement value and the detection value of the vibration environment, and the training samples are constructed by combining the standard data of the gravimeter in a vibration-free environment;

[0091] Step 5, construct a BP neural network, set the input layer node of the BP neural network to 2, and the output layer node to 1. Select a single hidden layer network structure and set the number of nodes to 6 based on experience, input the training samples into the BP neural network, train the BP neural network model with the training samples, and use the obtained model to compensate the residual errors outside the vibration compensation optimization model again to obtain the final vibration compensation result of the gravimeter.

[0092] Example 2

[0093] On the basis of Example 1, the specific process of obtaining the optimal delay coefficient and gain coefficient in the optimization model through the ADMM algorithm improved by the neighbor term is as follows:

[0094] Step 3.1, initialize the parameters in the ADMM algorithm: constant penalty factor σ∈(0,1), vector update coefficient β∈[0.5,1), initial multiplier variable λ0 is a column vector with all elements set to 1, golden ratio Γ and K are the feasible domains of τ and k respectively, and τ0∈Γ, k0∈K, the neighboring term coefficients S and T are set to random numbers in (0, 0.5), and the maximum number of iterations of the algorithm is set;

[0095] Step 3.2: Update the parameters of the τ-subproblem based on the improvement of the golden ratio neighboring term. At this moment, only the τ-subproblem is in the calculation state. After the update of this problem is completed, it will be shared with the k-subproblem. The update formula is:

[0096]

[0097]

[0098] Where the subscript n is the update algebra, u represents the process variable τ is golden sectioned and u0=τ0,η n is the update weight of the neighboring items of τ, y represents the variable to be calculated;

[0099] Step 3.3: Half update of the column vector of multiplier variables; the update formula is:

[0100]

[0101] Step 3.4: Update the parameters of the k-subproblem according to the algorithm principle. After the calculation is completed, share it with another subproblem for the next round of optimization calculation until the maximum number of iterations is reached. The parameter update process is as follows:

[0102]

[0103] where μ n is the update weight of k neighboring items;

[0104] Step 3.5: Update the calculation of the multiplier variables and other process variables. After reaching the maximum number of updates, the global optimal model coefficient gain coefficient k is obtained. opt and the delay factor τ opt ; The update formula is:

[0105]

[0106]

[0107] ξ n+1 =ρμ n+1 .

[0108] Example 3

[0109] The gravimeter was placed on a vibration table and subjected to a 20-min static test and an irregular random vibration test. Figure 3 The experimental results show that the accuracy of gravity measurement is significantly affected by the vibration environment. The evolution of gravity measurement accuracy after iterative compensation is as follows: Figure 4 As shown in Figure 1, it can be seen that with the increase in the number of iterations, the accuracy gradually improves. The final vibration compensation result is shown in Figure 1. Figure 5 The gravity measurement accuracy is shown in Table 1:

[0110] Table 1

[0111]

[0112] Vibration compensation test results Figure 3 , Figure 4 , Figure 5 Combined with Table 1, it can be seen that although the accuracy of the gravity measurement results obtained after vibration compensation by the method of the present invention is not as good as the measurement accuracy in a vibration-free environment, it effectively suppresses the vibration error and significantly improves the gravity measurement accuracy in a vibration environment, which has important reference value for improving the environmental adaptability of high-precision gravity measurement in the field.

[0113] Through the above-mentioned method, the vibration compensation method of the zero-length spring relative gravimeter of the present invention improves the optimization effect and convergence speed by using the ADMM algorithm improved by the neighboring term, and proposes to perform vibration compensation by iterative calculation, which effectively avoids the problem of insufficient or overcorrection of vibration compensation in a single calculation, and improves the accuracy of vibration compensation. In combination with the BP neural network, a compensation model of non-modeled errors is constructed to compensate the residual errors after the initial compensation again, which fully overcomes the disadvantage that gravity measurement is significantly affected by the vibration environment, greatly improves the accuracy of gravity measurement in a vibration environment, and can provide technical support for improving the environmental adaptability of high-precision gravity measurement in the field.

Claims

1. A gravity meter vibration compensation method using iterative calculation and BP neural network, characterized in that: Follow these steps to implement: Step 1, obtaining preliminary gravity measurement values ​​of a zero-length spring relative gravimeter and a flexible accelerometer gravimeter and detection values ​​of a vibration environment, and filtering the detection values ​​of the vibration environment to obtain vibration acceleration; Step 2: Establish a vibration compensation optimization model through vibration transfer relationship and ADMM algorithm; Step 3: Obtain the optimal delay coefficient and gain coefficient in the optimization model through the ADMM algorithm improved by the neighboring term; Step 4: Compensate the initial gravity measurement value and update the detection value of the vibration environment. Through iterative calculation, the optimal compensated gravity measurement value and the iteratively updated vibration environment detection value are obtained, and the training sample is constructed in combination with the standard data of the gravimeter in a vibration-free environment. Step 5: Construct a BP neural network, and train the BP neural network model with training samples. The obtained model is used to compensate the residual errors outside the vibration compensation optimization model again to obtain the final vibration compensation result of the gravimeter.

2. The gravimeter vibration compensation method using iterative calculation and BP neural network in combination according to claim 1, characterized in that: The process of obtaining the preliminary gravity measurement value and the vibration environment detection value of the zero-length spring relative gravimeter and the flexible accelerometer gravimeter in step 1 is as follows: fix the zero-length spring relative gravimeter and the flexible accelerometer gravimeter and perform static measurement to obtain the preliminary gravity measurement value and the vibration environment detection value.

3. The gravimeter vibration compensation method using iterative calculation and BP neural network in combination according to claim 1, characterized in that: The filtering of the detection value of the vibration environment in step 1 is specifically as follows: a low-pass filter with a cutoff frequency of 0.05 Hz is designed to filter the detection value of the vibration environment to obtain vibration acceleration data a in the same frequency band as the preliminary measurement value of gravity. f .

4. The gravimeter vibration compensation method using iterative calculation and BP neural network in combination according to claim 1, characterized in that: The specific process of step 2 is as follows: Step 2.1: Use the delay gain model to simplify the real transfer function. The gain coefficient and delay coefficient remain unchanged during the current measurement process. Then: a c =k·a f (t+τ); where a c is the vibration compensation signal used to approximate the actual vibration condition of the mass block, t is the time mark of the disturbance acceleration measurement signal, k is the gain coefficient, and τ is the delay coefficient; Step 2.2: According to the basic principle of ADMM algorithm, the basic structure and augmented Lagrangian function of the vibration compensation optimization model constructed based on the delay gain model are: f(k)=std(g m -and c (k)); where g m is the measured value of gravity, g r is the true value of gravity, λ is the column vector of multiplier variables; σ is a constant penalty factor, ||·||2 is the vector two-norm operator, h(τ) and f(k) are the objective functions of the delay coefficient τ and the gain coefficient k respectively, corr is the correlation calculation function, and std is the standard deviation calculation function.

5. The gravimeter vibration compensation method using iterative calculation and BP neural network in combination according to claim 2, characterized in that: The specific process of step 3 is as follows: Step 3.1, initialize the parameters in the ADMM algorithm: constant penalty factor σ∈(0,1), vector update coefficient β∈[0.5,1), initial multiplier variable λ0 is a column vector with all elements set to 1, golden ratio Γ and K are the feasible domains of τ and k respectively, and τ0∈Γ, k0∈K, the neighboring term coefficients S and T are set to random numbers in (0, 0.5), and the maximum number of iterations of the algorithm is set; Step 3.2: Update the parameters of the τ-subproblem based on the improvement of the golden ratio neighboring term. At this moment, only the τ-subproblem is in the calculation state. After the update of this problem is completed, it will be shared with the k-subproblem. The update formula is: Where the subscript n is the update algebra, u represents the process variable τ is golden sectioned and u0=τ0,η n is the update weight of the neighboring items of τ, y represents the variable to be calculated; Step 3.3: Half update of the column vector of multiplier variables; the update formula is: Step 3.4: Update the parameters of the k-subproblem according to the algorithm principle. After the calculation is completed, share it with another subproblem for the next round of optimization calculation until the maximum number of iterations is reached. The parameter update process is as follows: where μ n is the update weight of k neighboring items; Step 3.5: Update the calculation of the multiplier variables and other process variables. After reaching the maximum number of updates, the global optimal model coefficient gain coefficient k is obtained. opt and the delay factor τ opt ; The update formula is: x n+1 =rm n+1 。 6. The gravimeter vibration compensation method using iterative calculation and BP neural network in combination according to claim 5, characterized in that: The step 4 described in which the initial gravity measurement value is compensated and the detection value of the vibration environment is updated is to perform iterative calculation on the initial gravity measurement value and the detection value of the vibration environment by optimizing the optimal delay coefficient and gain coefficient in the model. The specific process is: a c-opt =k opt ·a f (t+τ opt ); g m =g m -a c-opt ; a f =a f -a c-opt 。 7. The gravimeter vibration compensation method using iterative calculation and BP neural network in combination according to claim 1, characterized in that: The iteration termination condition described in step 4 is: The difference in accuracy between two consecutive updated preliminary gravity measurements is less than 1μGal or reaches the maximum number of iterations of the algorithm.

8. The gravimeter vibration compensation method using iterative calculation and BP neural network in combination according to claim 1, characterized in that: In step 5, the input layer nodes of the BP neural network are 2, the output layer is 1 node, a single hidden layer network structure is selected and the number of nodes is set to 6.

Citation Information

Cited By

  • Vehicle-mounted gravity dynamic error compensation method and device based on intelligent algorithm

    CN120296332A

  • Motor motion curve optimization method and device based on iterative learning

    CN122386728A

  • Elastic element high-temperature testing device, design method and vibration compensation method

    CN122651319A