CN1888384A - Well drilling slope level azimuth angle error compensating method based on neural network - Google Patents

Abstract

A method for compensating the azimuth angle error of a drilling inclinometer based on a neural network includes the following four basic steps: 1) determining the structural model of the neural network according to the characteristics of the output signal of the azimuth angle of the drilling inclinometer; 2) obtaining the neural network training 3) Perform neural network training to obtain optimal neural network model parameters; 4) Calculate the real azimuth angle according to the inclination angle, the initially calculated azimuth angle, and the tool face angle. Neural network has the characteristics of distributed parallel processing, nonlinear mapping, robust fault tolerance and strong generalization ability, which makes it widely used in intelligent information processing. The trained neural network can approach the nonlinear relationship between the input and output of the drilling inclinometer system with high precision and has strong generalization ability, and can compensate the systematic error of the drilling inclinometer in the whole measurement range. The invention corrects the calculated azimuth angle, overcomes the system measurement errors caused by different azimuth angles, inclination angles and tool face angles, and outputs accurate azimuth angles. The method of the invention can also be used for modeling and compensation of output signals of other sensors.

Description

A Neural Network Based Azimuth Error Compensation Method for Drilling Inclinometer

technical field

The invention relates to a drilling inclinometer, in particular to a system error compensation method of the drilling inclinometer, which belongs to the application field of inertia technology and is suitable for directional drilling, well trace measurement and the like.

Background technique

The gyro inclinometer is an instrument that can be used for sidetracking and measuring well tracks in oilfield production. The window is actually opened in the direction of the oil layer, so that the old wells and old wells can be redeveloped, which not only saves expenses, but also greatly saves manpower and material resources. The currently widely used drilling inclinometer is composed of a two-degree-of-freedom flexible gyroscope and two quartz accelerometers. It adopts the working mode of the gyro compass and calculates the borehole inclination angle and tool face angle from the output of the accelerometer. On this basis, the azimuth angle can be calculated according to the output of the gyroscope. Before forming the drilling inclinometer, the gyroscope and accelerometer are modeled, tested and calibrated to improve the measurement accuracy of the inertial device and thus improve the accuracy of the output inclination angle, tool face angle and azimuth angle of the drilling inclinometer. In practical application, a simpler inertial device model is used for the convenience of engineering realization, ignoring the high-order items in the output error of the inertial device, and the installation error will be introduced when the inclinometer is installed, which will cause the output inclination angle of the inclinometer, Systematic error of tool face angle and azimuth. Especially when the inclination angle, azimuth angle and tool face angle change, the calculation error of the azimuth angle is different. After the mechanical installation and electrical debugging of the inclinometer are completed, the system error of this part is basically determined, and the repeatability is good. In order to compensate this part of the system error, the current method usually adopts the look-up table method, that is, the inclinometer system error table is established through experiments in advance, the actual output data is compared with the data in the table, and the system error corresponding to the output data at this time is obtained, and then The system error obtained by looking up the table is compensated. At present, the method of establishing the error table of the inclinometer system is: fix the gyro inclinometer on the position turntable through the fixture, and fix the tool face angle. In the case of a certain inclination angle, rotate the azimuth angle, and select some azimuth angles evenly within 360 degrees as the test points according to the accuracy requirements, and obtain the azimuth error of different test points, and repeat it at different inclination angles. above test. Finally, the corresponding relationship between inclination angle, azimuth angle and azimuth angle error of the inclinometer is established to form a table. When establishing the error table, some discrete test points are selected, and the method of interpolation is used for actual compensation. Selecting too many test points will increase the time to build the table, and selecting too few test points will affect the accuracy of compensation. The table look-up method currently used ignores the impact of the tool face angle on the azimuth angle. If the inclination angle, azimuth angle and tool face angle of the inclinometer change simultaneously to create a table for the output of the azimuth angle, the first is the workload of creating the table. It will be very large. Second, it is difficult to realize the interpolation algorithm in the case of three variables, and the compensation accuracy is not high.

Contents of the invention

Purpose of the present invention: overcome the deficiencies in the prior art, propose a kind of compensation method of azimuth error of drilling inclinometer based on neural network, realize the compensation of azimuth error of drilling inclinometer in the whole measuring range, output is accurate and high Accuracy of drilling inclinometer azimuth.

The technical solution of the present invention is: a method for compensating the azimuth angle error of a drilling inclinometer based on a neural network, which is characterized in that: using the nonlinear mapping function of the neural network, a deep forward neural network with three inputs and one output is selected to establish The mapping relationship between the inclination angle, tool face angle and preliminary calculated azimuth angle output by the drilling inclinometer and the real azimuth angle, the main realization steps are as follows:

(1) Determine the structural model of the neural network according to the characteristics of the azimuth output signal of the drilling inclinometer. When establishing the neural network model, the inclination angle, the tool face angle and the initially calculated azimuth angle output by the drilling inclinometer are used as the neural network model. The input volume, the real azimuth angle is the output volume to construct a three-input single-output neural network system, that is, select a forward neural network with a four-layer structure composed of the input layer, hidden layer 1, hidden layer 2 and output layer;

The neural network theory proves that a three-layer forward network can be used to approximate any nonlinear mapping, but for more complex functions, a deep forward network will achieve better results. Therefore, the present invention selects a forward neural network with four layers (ie input layer, hidden layer 1, hidden layer 2, output layer) structure. For the drilling inclinometer, the inclination angle, azimuth angle and tool face angle of the drilling inclinometer all have an influence on the calculation of the azimuth angle, which will cause an error in the calculation of the azimuth angle. Therefore, the initially calculated inclination angle, azimuth angle, and tool face angle are selected as the input of the neural network, and the corrected and compensated azimuth angle is used as the output of the neural network, that is, the number of nodes in the input layer of the neural network is 3, and the number of nodes in the output layer is 1 , the number of nodes in hidden layer 1 can be selected from 4 to 8, and the number of nodes in hidden layer 2 can be selected from 4 to 8.

(2) Obtain the training samples of the neural network to simulate the inclination angle, tool face angle and azimuth angle of the drilling inclinometer on the three-axis position turntable, and simulate it by rotating the three-axis position turntable around its three rotation axes The inclination angle, tool face angle and azimuth angle of the drilling inclinometer, and the raw data output by the inclinometer are collected at the same time, and the inclination angle and the initial calculation azimuth angle are calculated to form the learning samples of the neural network.

When using artificial neural networks to solve practical problems, the main work is to collect sample data. In the present invention, the acquisition of learning samples is carried out on a three-axis position turntable. The inclination angle, tool face angle and azimuth angle of the drilling inclinometer are simulated by the rotation of the three-axis position turntable around its three rotation axes. The test point selection rules are:

The selection rule of the inclination angle test point is to select within the range of 0 to 75 degrees. When the angle is close to zero, the sampling points are dense, and when the angle is close to 75 degrees, the interval of test sampling points is increased. For example, the interval taken within 10 degrees is small, greater than 10 degree is selected at an interval of 10 degrees; the selection principle of the test point for the azimuth angle is uniformly selected within the range of 0-360 degrees, and the sampling interval is selected according to the requirements of accuracy and experiment time; the selection principle for the test point of the tool face angle is 0 It is uniformly selected in the range of ~360 degrees, and the sampling interval is selected according to the requirements of precision and experiment time.

(3) Carry out neural network training to obtain optimal neural network model parameters.

Using the neural network model obtained in step (1) and the learning samples obtained in step (2) to train the neural network and obtain optimal weights.

(4) Calculate the real azimuth angle according to the inclination angle, the azimuth angle calculated initially, and the tool face angle.

The principle of the present invention is: the present invention is aimed at the drilling inclinometer adopting the gyrocompass method, and the inertial measurement part is composed of a dual-axis power tuning rate gyroscope, two quartz accelerometers and corresponding electronic circuits. Among them, the gyro is used to sense the earth's angular velocity component, and the acceleration is used to sense the gravity component. The azimuth, inclination and tool face angle of the drilling inclinometer, that is, the well pipe, are calculated by the output values of the gyro and acceleration. The geographic coordinate system XYZ (northeast sky) and the inclinometer coordinate system xyz are used to represent the azimuth, inclination angle and tool face angle of the well pipe, where the x-axis and y-axis are the sensitive axes of the two-axis gyroscope and the two-axis accelerometer . The coordinate system xyz of the inclinometer is consistent with the attitude of the measured well pipe, so the attitude of the xyz coordinate system relative to the XYZ coordinate system is the attitude of the well pipe, and thus the trajectory of the well pipe is obtained. The relationship between the xyz coordinate system and the XYZ coordinate system is shown in Figure 3. XYZ is the coordinate axis of the Northeast Heaven geographic coordinate system, and the xyz inclinometer coordinate system coordinate axis. x ₁ y ₁ z ₁ and x ₂ y ₂ z ₂ are coordinate systems in the process of coordinate system conversion. In Fig. 3, A is the azimuth angle, I is the inclination angle, T is the tool face angle, and A ^& , I ^& , T ^& are the corresponding angular velocity vectors respectively. The sequence of coordinate system rotation is: first rotate clockwise to the coordinate system x ₁ y ₁ z ₁ according to the Z axis, the rotation angle is the azimuth A, and then rotate the inclination angle I according to the y ₁ axis of the coordinate system x ₁ y ₁ z ₁ to the coordinate system x ₂ y ₂ z ₂ and finally rotate the inclinometer by the z ₂ axis of the x ₂ y ₂ z ₂ coordinate system to the coordinate system xyz

The projection components of the earth's rotation angular velocity and gravitational acceleration in the northeast sky coordinate system are:

ω＝[0 Ωcosφ Ωsinφ] ^T (1)

a=[0 0 g] ^T (2)

In the formula:

Ω is the angular velocity of the earth's rotation, φ is the latitude of the measuring point, and g is the gravitational acceleration of the measuring point.

After the coordinate system is rotated, the angular velocity and acceleration components in the inclinometer coordinate system are:

$\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}\end{array}\right]{<span\; class="notranslate">\; =</span>\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; \&Omega;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; T</span>}\\ \mathrm{<span\; class="notranslate">\; \&Omega;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; T</span>}\\ \mathrm{<span\; class="notranslate">\; \&Omega;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; I</span>}\end{array}\right]}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

$\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}\\ {\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}\\ {\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}\end{array}\right]<span\; class="notranslate">\; =</span>{\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; T</span>}\\ <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; T</span>}\\ <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; I</span>}\end{array}\right]}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

From the equations (1), (2), (3) and (4), the tool face angle T, inclination angle I and azimuth angle A can be calculated as:

$\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; arctan</span>}\frac{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arcsin</span>}\frac{\sqrt{{{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; +</span>{{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}}{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arctan</span>}\frac{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; I</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; I</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

With the accuracy of the current acceleration, the errors of the tool face angle T and the inclination angle I calculated by it are relatively small. The calculation formula of the azimuth angle includes the tool face angle T, the inclination angle I and the angular velocity components on the x-axis and the y-axis. The calculation error of the azimuth angle is the result of their joint action. When the gyro has errors and the system has installation errors, the azimuth angle calculation errors of the inclinometer are different under different attitudes, and have good repeatability. Therefore, it is possible to improve the accuracy of the azimuth angle by finding the error model through experiments in advance and compensating the azimuth angle of the inclinometer. It can be seen from formula (7) that when the inclination angle I is close to 90 degrees, the calculation error of the system will be relatively large. Therefore, when the gyro compass method is used, the inclination angle I is limited. Generally, the inclination angle I is less than 75 degrees.

The advantage of the present invention compared with prior art is:

(1) The present invention overcomes the shortcoming that the accuracy of the interpolation algorithm decreases when the inclination angle, the initially calculated azimuth, and the tool face angle are used as input variables in the table-building method, and realizes the inclination angle of the drilling inclinometer, and the azimuth angle of the initial calculation. , The mapping from the tool face angle to the azimuth angle achieves the effect of compensating the azimuth angle error.

(2) The subsection selection method of the inclination angle test point fully reflects the characteristics of the calculation error of the azimuth angle of the drilling inclinometer, and the learning samples formed are more targeted, which not only speeds up the training speed of the neural network, but also improves its calculation accuracy . In the case of achieving the same precision, the application of neural network method can take fewer test points than the look-up table method, thus improving the efficiency.

Description of drawings

Fig. 1 is the schematic diagram of neural network of the present invention;

Fig. 2 is the flow chart that the present invention obtains neural network learning sample;

Fig. 3 is a position relation diagram between the coordinate system of the inclinometer of the present invention and the geographic system.

Detailed ways:

For the technical scheme proposed above, take the following steps to realize the present invention:

(1) Establish a neural network model

The present invention takes the inclination angle of the drilling inclinometer, the tool face angle and the azimuth angle of preliminary calculation as the input variable of the neural network, and the real azimuth angle is used as the output variable of the neural network, and four layers (i.e. input layer, hidden layer 1, hidden layer 1, hidden layer 1) are selected. Layer 2, output layer) structure of the forward neural network. The number of nodes in the input layer of the neural network is 3, and the number of nodes in the output layer is 1. Considering that the influence of inclination and azimuth of the drilling inclinometer on the azimuth is more complicated, the number of hidden layer 1 nodes is selected as 8, and the number of hidden layer 2 nodes is selected as 8. The neural network structure is shown in Figure 1, where X1, X2, X3 is the input variable of the neural network corresponding to the inclination angle, the tool face angle and the initially calculated azimuth angle output by the drilling inclinometer, and Y corresponds to the corrected azimuth angle, which is the output variable of the neural network.

(2) Determine the test points of the inclination angle and azimuth angle of the drilling inclinometer

The selection of the test points of the drilling inclinometer should not only cover the measuring range of the drilling inclinometer, but also reflect the characteristics of the output data of the drilling inclinometer. The test points of the inclination angle are sampled within the range of 0-75 degrees, and the sampling points are 3 degrees, 5 degrees, 10 degrees, 15 degrees, 25 degrees, 35 degrees, 45 degrees, 55 degrees, 65 degrees, and 75 degrees. The azimuth test points are uniformly sampled within the range of 0 to 360 degrees, and the sampling interval depends on the accuracy requirements. Because the neural network has the generalization ability, the sampling interval can be relaxed compared with the look-up table method, and the sampling interval is 30 degrees in the present invention. The test points of the tool face angle are uniformly sampled within the range of 0 to 360 degrees, and the sampling interval is 30 degrees.

(3) Obtain neural network learning samples

At the test point determined in (2), the inclination angle output by the drilling inclinometer, the preliminary calculation azimuth angle and the tool face angle are collected as the input value of the learning sample, and the azimuth angle simulated by the three-axis position turntable is used as the expected output value of the learning sample. The specific operation is to first turn the inclination angle, azimuth angle, and tool face angle to zero, and then turn the inclination angle, azimuth angle, and tool face angle in sequence, and turn all the test points, and collect the output data of the inclinometer at the same time to form a measurement that covers the entire measurement. Range of learning samples for the neural network. The learning sample acquisition process is shown in Figure 2. The number of test points N _i (i=1, 2, 3) respectively represents the number of test points for inclination angle, azimuth angle, and tool face angle, which can be determined according to the system accuracy requirements.

(4) Perform neural network training to obtain optimal neural network model parameters

On the basis of the neural network model and learning samples determined in the previous steps (1), (2), and (3), the network is trained using the BP algorithm. In the training process, first a set of model parameters is given, and the output of the neural network is calculated with these parameters, and then the output of the neural network is compared with the real value to obtain the calculation error. Then according to the error, modify the model parameters according to the BP algorithm, so that the network will continue to develop in the direction of correct response until the difference between the output of the network and the expected output is within the allowable range. The model parameters at this time are the optimal network model parameters, that is, the connection weights between neurons and the thresholds of neurons. The weight modification process is as follows.

Define the error function as $e_p=\frac{1}{2}{(t_p-{\mathrm{the\; y}}_p)}^2,$ Among them, t _p is the expected output value of the neural network, and y _p is the calculated value of the neural network. according to $\mathrm{\&Delta;W}=-\mathrm{\&eta;}\frac{{\mathrm{\&delta;\; e}}_p}{\mathrm{\&delta;W}}$ To modify the connection weights between neurons, and finally reach the minimum e _p , and determine that the parameter is the optimal parameter at this time.

(5) Calculate the true azimuth angle according to the inclination angle, the tool face angle and the initially calculated azimuth angle by $A=\arctan \frac{{\mathrm{\&omega;}}_x\cos T-{\mathrm{\&omega;}}_{\mathrm{the\; y}}\sin T+\mathrm{\&Omega;}\sin \mathrm{\&phi;}\sin I}{({\mathrm{\&omega;}}_x\sin T+{\mathrm{\&omega;}}_{\mathrm{the\; y}}\cos T)\cos I}$ It can be seen from the formula that the calculation formula of the azimuth angle includes the tool face angle T, the inclination angle I and the angular velocity components on the x-axis and the y-axis, and the calculation error of the azimuth angle is the result of their joint action. When the gyro has errors and the system has installation errors, the azimuth angle calculation errors of the inclinometer are different under different attitudes, and have good repeatability.

In step (4), the optimal parameters are obtained, that is, the nonlinear mapping relationship between the initially calculated tool face angle, inclination angle, azimuth angle and the real azimuth angle is established, that is, the initially calculated Tool face angle, inclination angle, azimuth angle, the neural network will output the real azimuth angle.

In computation, the output of neurons in the previous layer serves as the input to the neurons in the next layer. Suppose the input signal of the neural network: the initially calculated tool face angle, inclination angle and azimuth angle are x ₁ , x ₂ , x ₃ respectively, and the output signal is y. The following is the process of finding y from x ₁ , x ₂ , and x ₃ .

The input of the first hidden layer neuron is:

${\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; L\; m</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

In the above formula, n is the number of input neurons, here n=3, m1 is the number of input neurons, here m1=8, w1 _{i, j} is the distance between the first hidden layer neuron i and the input neuron The connection weight of , θ1 _j is the threshold of the first hidden layer neuron.

The output of the neurons in the first hidden layer is:

O1 _i ＝f(I1 _i ) (9)

There are m2 neurons in the second hidden layer, and the input of each neuron is:

${\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; L\; m</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

In the above formula, m2=8, w2 _{i, j} are the connection weights between the neurons of the second hidden layer and the neurons of the first hidden layer, and θ2 _j is the threshold of the neurons of the second hidden layer.

The output of the neurons in the second hidden layer is:

O2 _i ＝f(I2 _i ) (11)

Take the threshold value of the output neuron as zero, and take the linear function as the activation function of the neuron in the output layer, then the output of the output neuron (that is, the output of the entire network) is:

$\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 12</span>}<span\; class="notranslate">\; )</span>$

In the above formula, v _i is the connection weight between the output neuron and the second hidden layer neuron, and y calculated by formula (12) is the real azimuth angle after compensation.

Claims (3)

1, a kind of drilling inclinometer azimuth error compensation method based on neural network, it is characterized in that: utilize the nonlinear mapping function of neural network, select the deep layer forward neural network of three input single output, set up the output of drilling inclinometer The mapping relationship between the inclination angle, the tool face angle and the initially calculated azimuth angle and the real azimuth angle, the main realization steps are as follows: (1) Determine the structural model of the neural network according to the characteristics of the azimuth output signal of the drilling inclinometer. When establishing the neural network model, the inclination angle, the tool face angle and the initially calculated azimuth angle output by the drilling inclinometer are used as the neural network model. The input volume, the real azimuth angle is the output volume to construct a three-input single-output neural network system, that is, select a forward neural network with a four-layer structure composed of the input layer, hidden layer 1, hidden layer 2 and output layer; (2) Obtain the training samples of the neural network to simulate the inclination angle, tool face angle and azimuth angle of the drilling inclinometer on the three-axis position turntable, and simulate it by rotating the three-axis position turntable around its three rotation axes The inclination angle, tool face angle and azimuth angle of the drilling inclinometer; (3) Neural network training is carried out to obtain optimal neural network model parameters; (4) Calculate the real azimuth angle according to the inclination angle, the azimuth angle calculated initially, and the tool face angle. 2, a kind of neural network-based drilling inclinometer azimuth error compensation method according to claim 1, is characterized in that: the input layer node of the forward neural network that selects four-layer structure in the described step (1) is 3. Nodes in the hidden layer 1 are 4 to 8, nodes in the hidden layer 2 are 4 to 8, and nodes in the output layer are 1. 3. A method for compensating the azimuth angle error of the drilling inclinometer based on neural network according to claim 1, characterized in that: the inclination angle, the tool face angle and the azimuth angle of the drilling inclinometer in the step (2) The selection rules of the test points are: the test points of the inclination angle are sampled within the range of 0-75 degrees; the test points of the azimuth angle are sampled uniformly within the range of 0-360 degrees; sampling.