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

Abstract

The present invention relates to a borehole clinometer azimuth angle error compensation method based on neural network. Said method includes the following four steps: 1), according to the characteristics of borehole clinometer azimuth angle output signal defining structure model of neural network; 2), obtaining training sample of neural network; 3), making neural network training and obtaining optimum neural network model parameter; and 4), utilizing inclination angle, primarily-calculated azimuth angle and tool face angle to calculate real azimuth angle.

Description

A kind of well drilling slope level azimuth angle error compensating method based on neutral net

Technical field

The present invention relates to a kind of well drilling slope level, particularly a kind of well drilling slope level systematic error compensation method belongs to the inertial technology application, is applicable to directed drilling, the measurement of well mark etc.

Background technology

Gyrolevel is exactly a kind of instrument that can play window sidetracking, measuring well mark in field produces, this instrument can go out orientation to be windowed by gyro to measure in the down-hole, like this can be in Jiu Jing, old down-hole by actual the windowing of actual oil reservoir direction, make Jiu Jing, Lao Jing obtain secondary development, not only save spending, and saved manpower and materials greatly.Current application well drilling slope level comparatively widely is made up of flexible gyroscope and two quartz accelerometers of a double freedom, adopt the working method of gyrocompass, output by accelerometer calculates wellhole angle of slope and tool face azimuth, on this basis, can calculate the azimuth according to the output of gyro again.Before forming well drilling slope level, gyro and accelerometer are all carried out modeling, test and staking-out work, thereby the certainty of measurement that improves inertia device improves well drilling slope level output angle of slope, tool face azimuth and azimuthal precision.Be that Project Realization has conveniently adopted better simply inertia device model when practical application, ignored the higher order term in the inertia device output error, can introduce alignment error when inclinometer is installed, these all can cause inclinometer output angle of slope, tool face azimuth and azimuthal systematic error.Particularly at the angle of slope, when azimuth and tool face azimuth change, azimuthal error of calculation is different.This part systematic error is definite substantially after the machinery installation of inclinometer is finished with electric adjustment, and repeatability better.For compensating this part systematic error, method adopts look-up table usually at present, promptly sets up inclinometer systematic error table by test in advance, data comparison in actual output data and the table, obtain the systematic error of output data correspondence this moment, the systematic error compensation that obtains tabling look-up falls then.At present, the method for inclinometer systematic error table foundation is: by frock clamp gyrolevel is fixed on the position turntable, the tool face azimuth is fixed.Constant situation under a certain angle of slope, some azimuths are evenly chosen as test point according to required precision in the rotational orientation angle in 360 degree scopes, draw the azimuth angle error of different test points, repeat above test under different inclination angle.Finally set up the corresponding relation of inclinometer angle of slope, azimuth and azimuth angle error, form form.Be to choose some discrete test points when setting up the error form, adopt the method for interpolation during actual the compensation.Test point is chosen too much can increase the time of building table, and test point is chosen the precision that can influence compensation very little again.The look-up table that adopts has been ignored the tool face azimuth to azimuthal influence at present, if form is set up in azimuthal output under the situation that inclinometer angle of slope, azimuth and tool face azimuth are changed simultaneously, the one, the workload of setting up form will be very big, the 2nd, the realization of interpolation algorithm has certain difficulty under the situation of ternary, and compensation precision is not high.

Summary of the invention

Purpose of the present invention: overcome the deficiencies in the prior art, a kind of well drilling slope level azimuth angle error compensating method based on neutral net is proposed, in whole measurement category, realize the well drilling slope level azimuth angle compensation of error is exported accurate and high-precision well drilling slope level azimuth angle.

Technical solution of the present invention is: a kind of well drilling slope level azimuth angle error compensating method based on neutral net, it is characterized in that: the Nonlinear Mapping function of utilizing neutral net, select the deep layer feedforward neural network of the single output of three inputs, the azimuth of angle of slope, tool face azimuth and primary Calculation of setting up well drilling slope level output is with the mapping relations between the true bearing angle, and main performing step has:

(1) determines the structural model of neutral net according to the characteristics of well drilling slope level azimuth angle output signal, the azimuth of angle of slope, tool face azimuth and the primary Calculation of exporting with well drilling slope level when setting up neural network model is as the input quantity of neutral net, the true bearing angle is the nerve network system that output quantity makes up the single output of three inputs, promptly selects the feedforward neural network of the four-layer structure be made up of input layer, hidden layer 1, hidden layer 2 and output layer;

Neural network theory proof can be approached any one Nonlinear Mapping with three layers of feedforward network, but can obtain better effect for the function of more complicated during with the deep layer feedforward network.Therefore, the present invention selects the feedforward neural network of four layers of (being input layer, hidden layer 1, hidden layer 2, output layer) structure.For well drilling slope level, the angle of slope of well drilling slope level, azimuth and tool face azimuth are all influential to azimuthal calculating, can produce the azimuth and calculate error.Therefore, select primary Calculation angle of slope, azimuth and tool face azimuth as the input of neutral net, azimuth behind the correction-compensation is as the output of neutral net, be that neutral net input layer number is 3, output layer node number is 1, hidden layer 1 node number is chosen as 4～8, and hidden layer 2 node numbers are chosen as 4～8.

(2) obtain the training sample of neutral net, carry out on three shaft position turntables at angle of slope, tool face azimuth and the azimuth of simulation well drilling slope level, simulate angle of slope, tool face azimuth and the azimuth of well drilling slope level around the rotation of its three turning cylinders by three shaft position turntables, gather the initial data of inclinometer output simultaneously and calculate the angle of slope and the primary Calculation azimuth, form the learning sample of neutral net.

When adopting artificial neural network to solve practical problem, main work is to collect sample data.Obtaining on three shaft position turntables of learning sample carried out among the present invention.Simulate angle of slope, tool face azimuth and the azimuth of well drilling slope level around the rotation of its three turning cylinders by three shaft position turntables.The test point selection rule is:

The selection rule of angle of slope test point is to choose in 0～75 degree scope, and sample point is close when approaching zero, and angle is tested sample point constantly near 75 degree and increased at interval, as little with interior interval of getting at 10 degree, chooses with the interval of 10 degree greater than 10 degree; The selection principle of azimuthal test point is evenly to choose in 0～360 degree scope, and the sampling interval is chosen according to the requirement of precision and experimental period; The selection principle of the test point of tool face azimuth is evenly to choose in 0～360 degree scope, and the sampling interval is chosen according to the requirement of precision and experimental period.

(3) carry out neural metwork training, obtain optimum neural network model parameter.

The learning sample neural network training that neural network model that employing step (1) obtains and step (2) obtain also obtains optimum weights.

(4) calculate the true bearing angle according to azimuth, the tool face azimuth of angle of slope, primary Calculation.

Principle of the present invention is: the present invention is directed to the well drilling slope level that adopts the gyrocompass method, the inertia measurement part is made up of a double-shaft power tuning speed gyro, two quartz accelerometers and corresponding electronic circuit.Wherein gyro is used for responsive earth angular velocity component, acceleration and is used for responsive weight component.Calculate well drilling slope level just azimuth, angle of slope and the tool face azimuth of well casing by the output valve of gyro and acceleration.Adopt geographic coordinate system XYZ (sky, northeast) and inclinometer coordinate system xyz when the azimuth of representing well casing, angle of slope and tool face azimuth, wherein x axle, y axle are the sensitive axes of two axis gyroscope and two axis accelerometers.Inclinometer coordinate system xyz is consistent with the attitude of measuring well casing, so the attitude of the relative XYZ coordinate of xyz coordinate system system is the attitude of well casing, obtains the track of well casing thus.The relation of the relative XYZ coordinate of xyz coordinate system system as shown in Figure 3, XYZ is the geographical coordinate system coordinate axes in sky, northeast, xyz inclinometer coordinate system coordinate axes.x ₁y ₁z ₁And x ₂y ₂z ₂It is the coordinate system in the coordinate system transfer process.A is that azimuth, I are that angle of slope, T are the tool face azimuth among Fig. 3, A ^﹠amp;, I ^﹠amp;, T ^﹠amp;Be respectively corresponding angular velocity vector.The order of coordinate system rotation is: be rotated clockwise to coordinate system x by the Z axle earlier ₁y ₁z ₁,, the anglec of rotation is azimuth A, presses coordinate system x then ₁y ₁z ₁Y ₁Angle of slope I is to coordinate system x in the axle rotation ₂y ₂z ₂, press x at last ₂y ₂z ₂The z of coordinate system ₂Axle rotation inclinometer is to coordinate system xyz

Rotational-angular velocity of the earth and the acceleration of gravity projection components under day coordinate system northeastward are:

ω＝[0?Ωcosφ?Ωsinφ] ^T (1)

a＝[0?0?g] ^T (2)

In the formula:

Ω is a rotational-angular velocity of the earth, and φ is the latitude of survey mark, and g is the acceleration of gravity of survey mark.

Through after the coordinate system rotation, be at the angular velocity and the component of acceleration of inclinometer coordinate system:

$\left[\begin{array}{c}{\mathrm{\ω}}_x\\ {\mathrm{\ω}}_y\\ {\mathrm{\ω}}_z\end{array}\right]{=\left[\begin{array}{c}\mathrm{\Ω}\cos \mathrm{\φ}(\sin A\cos I\cos T+\cos A\sin T)-\mathrm{\Ω}\sin \mathrm{\φ}\sin I\cos T\\ \mathrm{\Ω}\cos \mathrm{\φ}(-\sin A\cos I\sin T+\cos A\cos T)+\mathrm{\Ω}\sin \mathrm{\φ}\sin I\sin T\\ \mathrm{\Ω}\cos \mathrm{\φ}\sin A\sin I+\mathrm{\Ω}\sin \mathrm{\φ}\cos I\end{array}\right]}^T---\left(3\right)$

$\left[\begin{array}{c}{a}_x\\ a_y\\ a_z\end{array}\right]={\left[\begin{array}{c}g\sin I\cos T\\ -g\sin I\sin T\\ -g\cos I\end{array}\right]}^T---\left(4\right)$

Can be regarded as to such an extent that tool face azimuth T, angle of slope I, azimuth A are by (1), (2), (3), (4) formula:

$T=-\arctan \frac{a_y}{a_x}---\left(5\right)$

$I=\arcsin \frac{\sqrt{{a^2}_x+{a^2}_y}}{g}---\left(6\right)$

$A=\arctan \frac{{\mathrm{\ω}}_x\cos T-{\mathrm{\ω}}_y\sin T+\mathrm{\Ω}\sin \mathrm{\φ}\sin I}{({\mathrm{\ω}}_x\sin T+{\mathrm{\ω}}_y\cos T)\cos I}---\left(7\right)$

With the precision of present acceleration, less by the error ratio of tool face azimuth T, the angle of slope I of its calculating.Comprised tool face azimuth T, angle of slope I and the angular velocity component on x axle y axle in azimuthal calculating formula, azimuthal error of calculation is their coefficient results.When gyro has the sum of errors system that alignment error is arranged, cause slope level azimuth angle error of calculation difference under the different attitudes, and good repeatability is arranged.Therefore, find the model of its error in advance by test, and the azimuth of inclinometer is compensated, can improve azimuthal precision.The system-computed error can be bigger when formula (7) is as seen near angle of slope I approaches 90 degree, therefore when adopting the gyrocompass mode angle of slope I limited, and generally gets angle of slope I and spend less than 75.

The present invention's advantage compared with prior art is:

(1) the present invention has overcome the shortcoming of interpolation algorithm precise decreasing when building in the table method the azimuth of angle of slope, primary Calculation, tool face azimuth as input variable, the azimuth, tool face azimuth of having realized well drilling slope level angle of slope, primary Calculation reach the effect to computer azimuth angle error compensation to azimuthal mapping.

(2) the segmentation system of selection of angle of slope test point has fully reflected the characteristics of well drilling slope level azimuth angle error of calculation, and the learning sample of formation is more targeted, has promptly accelerated the training speed of neutral net, has improved its computational accuracy again.Reaching under the situation of same precision, the comparable look-up table of Application of Neural Network Technology is taked number of test points still less, thereby has improved efficient.

Description of drawings

Fig. 1 is the schematic diagram of neutral net of the present invention;

Fig. 2 obtains the flow chart of neural network learning sample for the present invention;

Fig. 3 is the location diagram between inclinometer coordinate system of the present invention and Department of Geography.

The specific embodiment:

At technical scheme set forth above, take following step to realize the present invention:

(1) sets up neural network model

The present invention gets the azimuth of angle of slope, tool face azimuth and primary Calculation of well drilling slope level as the input variable of neutral net, the feedforward neural network of four layers of (being input layer, hidden layer 1, hidden layer 2, output layer) structure is chosen as the output variable of neutral net in the true bearing angle.Neutral net input layer number is 3, and output layer node number is 1.Consider that the inclination of well drilling slope level and orientation are to azimuthal more complicated that influences, hidden layer 1 node number elects 8 as, hidden layer 2 node numbers elect 8 as, neural network structure as shown in Figure 1, wherein X1, X2, X3 are the azimuths of angle of slope, tool face azimuth and primary Calculation of the respectively corresponding well drilling slope level output of input variable of neutral net, azimuth after the corresponding correction of Y is the output variable of neutral net.

(2) determine angle of slope and azimuthal test point of well drilling slope level

The well drilling slope level test point choose the characteristics that the measurement category that promptly will cover well drilling slope level reflects the well drilling slope level output data again.The test point at angle of slope is taken a sample in 0～75 degree scope, and sample point is 3 degree, 5 degree, 10 degree, 15 degree, 25 degree, 35 degree, 45 degree, 55 degree, 65 degree, 75 degree.The uniform sampling in 0～360 degree scope of azimuthal test point, decide on required precision the sampling interval.Because neutral net has generalization ability, the comparable look-up table in sampling interval relaxes, and the sampling interval is 30 degree among the present invention.The uniform sampling in 0～360 degree scope of the test point of tool face azimuth, sampling interval are 30 degree.

(3) obtain the neural network learning sample

(2) definite test point place gather angle of slope, primary Calculation azimuth and the tool face azimuth of well drilling slope level output input value as learning sample, angle, three shaft position turntable simulated-aximuths is as the desired output of learning sample.Concrete operations are earlier angle of slope, azimuth, tool face azimuth to be forwarded to zero-bit, order is rotated angle of slope, azimuth, tool face azimuth then, change all over whole test points, gather the inclinometer output data simultaneously, form the learning sample of the neutral net that covers whole measurement category.Learning sample obtains flow process as shown in Figure 2, number of checkpoints N _i(i=1,2,3) represent the number of checkpoints of angle of slope, azimuth, tool face azimuth respectively, can require to determine according to system accuracy.

(4) carry out neural metwork training, obtain optimum neural network model parameter

(1), (2), (3) go on foot on the basis of definite neural network model and learning sample in front, adopt the BP algorithm that network is trained.In training process, at first provide a group model parameter, with the output of this calculation of parameter neutral net, the output of neutral net and actual value are compared obtain error of calculation again.Then according to error, press the BP algorithm and revise model parameter, network is constantly developed, within the range of permission up to the difference of the output of the output of network and expectation towards the direction that can correctly respond.The model parameter of this moment is optimum network model parameter, just interneuronal connection weights and neuronic threshold value.The weights modification process is as follows.

The definition error function is $e_p=\frac{1}{2}{(t_p-y_p)}^2,$ T wherein _pBe the desired output of neutral net, y _pIt is the neutral net calculated value.Press $\mathrm{\ΔW}=-\mathrm{\η}\frac{{\mathrm{\δe}}_p}{\mathrm{\δW}}$ Go to revise and connect weights between neuron, finally reach e _pMinimum determines that this moment, parameter was an optimized parameter.

(5) according to the azimuth of angle of slope, tool face azimuth and primary Calculation calculate the true bearing angle by $A=\arctan \frac{{\mathrm{\ω}}_x\cos T-{\mathrm{\ω}}_y\sin T+\mathrm{\Ω}\sin \mathrm{\φ}\sin I}{({\mathrm{\ω}}_x\sin T+{\mathrm{\ω}}_y\cos T)\cos I}$ As seen formula has comprised tool face azimuth T, angle of slope I and the angular velocity component on x axle y axle in azimuthal calculating formula, azimuthal error of calculation is their coefficient results.When gyro has the sum of errors system that alignment error is arranged, cause slope level azimuth angle error of calculation difference under the different attitudes, and good repeatability is arranged.

In (4) step, obtained optimized parameter, just set up the Nonlinear Mapping relation between the tool face azimuth that primary Calculation goes out, angle of slope, azimuth and true bearing angle, be tool face azimuth, angle of slope, the azimuth that neuralward network input primary Calculation goes out, neutral net just can be exported the true bearing angle.

In calculating, the neuronic output of preceding one deck is as the neuronic input of one deck down.If neutral net input signal: the tool face azimuth that primary Calculation goes out, angle of slope, azimuth are respectively x ₁, x ₂, x ₃, output signal is y.Be by x below ₁, x ₂, x ₃Obtain the process of y.

First hidden neuron is input as:

${I1}_i=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{w1}_{i,j}{x}_j+{\mathrm{\θ}1}_j(i=\mathrm{1,2},\mathrm{L\; m}1)---\left(8\right)$

N is the number of input neuron in the following formula, and at this n=3, m1 is the number of input neuron, at this m1=8, w1 _{I, j}Be the first hidden neuron i with input neuron between be connected weights, θ 1 _jIt is the threshold value of first hidden neuron.

First hidden neuron is output as:

O1 _i＝f(I1 _i) (9)

Total m2 neuron in second hidden layer, each neuronic input is:

${I2}_i=\underset{j=1}{\overset{m1}{\mathrm{\Σ}}}{w2}_{i,j}{x}_j+{\mathrm{\θ}2}_j(i=\mathrm{1,2},\mathrm{L\; m}2)---\left(10\right)$

M2=8 in the following formula, w2 _{I, j}Be second hidden neuron with first hidden neuron between be connected weights, θ 2 _jIt is the threshold value of second hidden neuron.

Second hidden neuron is output as:

O2 _i＝f(I2 _i) (11)

The threshold value of getting output neuron is zero, and the line taking function is the neuronic excitation function of output layer, and the then output of output neuron (output of whole network just) is:

$y=\underset{i=1}{\overset{m2}{\mathrm{\Σ}}}{v}_i{O2}_i---\left(12\right)$

V in the following formula _iBe the be connected weights of output neuron with second hidden neuron, the y that is obtained by formula (12) is real azimuth, compensation back.

Claims (3)

1, a kind of well drilling slope level azimuth angle error compensating method based on neutral net, it is characterized in that: the Nonlinear Mapping function of utilizing neutral net, select the deep layer feedforward neural network of the single output of three inputs, the azimuth of angle of slope, tool face azimuth and primary Calculation of setting up well drilling slope level output is with the mapping relations between the true bearing angle, and main performing step has:

(1) determines the structural model of neutral net according to the characteristics of well drilling slope level azimuth angle output signal, the azimuth of angle of slope, tool face azimuth and the primary Calculation of exporting with well drilling slope level when setting up neural network model is as the input quantity of neutral net, the true bearing angle is the nerve network system that output quantity makes up the single output of three inputs, promptly selects the feedforward neural network of the four-layer structure be made up of input layer, hidden layer 1, hidden layer 2 and output layer;

(2) obtain the training sample of neutral net, carry out on three shaft position turntables at angle of slope, tool face azimuth and the azimuth of simulation well drilling slope level, simulates angle of slope, tool face azimuth and the azimuth of well drilling slope level around the rotation of its three turning cylinders by three shaft position turntables;

(3) carry out neural metwork training, obtain optimum neural network model parameter;

(4) calculate the true bearing angle according to azimuth, the tool face azimuth of angle of slope, primary Calculation.

2, a kind of well drilling slope level azimuth angle error compensating method according to claim 1 based on neutral net, it is characterized in that: selecting the input layer of the feedforward neural network of four-layer structure in the described step (1) is 3, hidden layer 1 node is 4～8, hidden layer 2 nodes are 4～8, and the output layer node is 1.

3, a kind of well drilling slope level azimuth angle error compensating method based on neutral net according to claim 1 is characterized in that: the selection rule of angle of slope, tool face azimuth and the azimuth test point of well drilling slope level is in the described step (2): the test point at angle of slope is taken a sample in 0～75 degree scope; The uniform sampling in 0～360 degree scope of azimuthal test point; The uniform sampling in 0～360 degree scope of the test point of tool face azimuth.