A real-time inversion method for azimuthal electromagnetic logging data while drilling

Abstract

The present invention discloses a method for real-time inversion of azimuthal electromagnetic wave logging data while drilling, which relates to the technical field of oil exploration and development, including: establishing various types of horizontal formation models, carrying out forward logging simulation, and obtaining a response signal library; designing an azimuthal electromagnetic wave logging while drilling inversion model based on the response signal library; inputting logging data, and automatically processing the data using a scaling and fidelity method; introducing a neural network algorithm, establishing a formation model integrated multi-initial value strategy, and optimizing initial values ​​for actual logging data; extracting the upper / lower interface position and resistivity distribution of the instrument in real time based on a regularized gradient optimization algorithm and initial values; statistically processing multiple groups of inversion results to obtain the final formation model and the credibility of each inversion parameter. The present invention can solve the problems of significant differences in electromagnetic wave logging data while drilling, drastic changes in the objective function, and many minimum values, and provides a constraint and optimization method for selecting initial values ​​for inversion.

Description

Technical Field

[0001] The invention relates to the technical field of petroleum exploration and development, and in particular to a method for real-time inversion of while-drilling azimuthal electromagnetic wave logging data. Background Art

[0002] Azimuth electromagnetic wave logging data while drilling has unique advantages such as large detection range and strong sensitivity to formation boundaries. It is widely used in the exploration and development of complex and unconventional oil and gas reservoirs. However, this type of logging data usually provides multiple types of detection modes, and the contribution of each type of data is difficult to control, resulting in a small proportion of data containing formation interface information and serious distortion of inversion results, which in turn leads to a decrease in geosteering capabilities. Therefore, how to perform fidelity processing on measured data, especially weak signals, to ensure the accuracy of inversion is one of the key issues to be solved in improving geosteering and reservoir evaluation.

[0003] At present, the inversion of azimuthal electromagnetic logging while drilling mainly focuses on the improvement of inversion algorithms, while there is little research on the optimization of logging data fidelity during the inversion process. From the perspective of inversion data, the real-time processing of azimuthal electromagnetic logging while drilling data mainly has the following problems: 1) There are multiple solutions and local convergence in the inversion, which makes it difficult to select the initial value; 2) The magnitude of various types of inversion data varies greatly, it is difficult to unify the scale, and the contribution ratio is difficult to control; 3) The geological signals in the inversion data change dramatically, resulting in distortion in the inversion process of the stratigraphic structure.

[0004] Therefore, it is urgent to study a real-time inversion method for azimuthal electromagnetic logging data while drilling. Summary of the invention

[0005] In order to provide real and effective logging inversion data for real-time geological guidance and reservoir evaluation, the present invention discloses a real-time inversion method for azimuthal electromagnetic wave logging data while drilling, which can solve the problems of significant differences in azimuthal electromagnetic wave logging data while drilling, drastic changes in objective functions and many minimum values, and provides constraints and optimization methods for selecting initial inversion values, thereby avoiding inversion iterations from falling into local minimum values ​​and reducing invalid initial models.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The real-time inversion method of azimuthal electromagnetic wave logging data while drilling includes the following steps:

[0008] s1. Establish horizontal formation models with different formation resistivity distribution, formation contrast, anisotropy, and well inclination angles, carry out forward simulation of azimuthal electromagnetic logging while drilling at each formation depth point, and obtain the azimuthal electromagnetic logging while drilling response signal library;

[0009] s2. Based on the response characteristics of apparent resistivity and geological signals in the response signal library, analyze the sensitivity of the response signal to different geological parameters, perform equivalent processing on complex three-dimensional models, and design an inversion model for azimuthal electromagnetic wave logging while drilling;

[0010] s3. Construct a scaling factor function, input a apparent resistivity curve and b geological signal curve, establish the reverse fitting relationship between the apparent resistivity and the amplitude ratio and phase difference signal, and convert the apparent resistivity curve into the amplitude ratio and phase difference signal;

[0011] s4. Eliminate the base value of the amplitude ratio signal, perform logarithmic processing on the amplitude ratio and phase difference signals, determine the normalization coefficients of the amplitude ratio, phase difference and geological signal, select the optimal scaling factor of the geological signal, and automatically process the data using the scaling fidelity method;

[0012] s5. Introduce neural network algorithm, combine shallow detection apparent resistivity value and prior information constraints, establish formation model integrated multi-initial value strategy, and optimize initial value based on actual logging data;

[0013] s6. Based on the regularized gradient optimization algorithm and the initial value of step s5, the upper / lower interface position and resistivity distribution of the instrument are extracted in real time;

[0014] s7. Perform statistical processing on multiple sets of inversion results to obtain the final formation model and the credibility of each inversion parameter.

[0015] Optionally, in step s4, the steps of the scaling fidelity processing method specifically include:

[0016] s4.1 For the amplitude ratio signal Att, simulate the amplitude base value A0 in the air medium, calculate Att-Att0 as the final amplitude ratio signal, and take the logarithm to get log 10 (Att_p), select log 10 The normalization coefficient K1 of (Att_p) is calculated Get normalized;

[0017] s4.2 For the phase difference signal Ps, similarly, take the logarithm of the phase difference signal to obtain log 10 (Ps), select log 10 The normalization coefficient K2 of (Ps) is calculated Get normalized;

[0018] s4.3 For the geological signal, set the original value of FLAG to -1, and determine whether the geological signal Gs is greater than zero. If it is greater than zero, assign the FLAG value to 1, otherwise execute step s4.4;

[0019] s4.4 Perform an absolute value operation on the Gs signal to obtain abs(Gs), and determine whether the abs(Gs) signal is less than the signal threshold. If so, assign the abs(Gs) signal to the signal threshold. Otherwise, execute step s4.5.

[0020] s4.5 Determine whether there is prior information of adjacent well data. If so, perform a forward simulation to calculate the geological signal Gs_bound1 at the interface based on the formation model S1 of the current layer resistivity Rt1 and the contrast CR1 in the prior information, and use Gs_bound1 to select the normalization coefficient K3, otherwise execute step s4.6;

[0021] s4.6 Based on the shallow detection apparent resistivity signal Ra as the instrument layer resistivity Rt2, with the formation contrast CR2 (ratio of surrounding rock resistivity to instrument layer resistivity) = 1:5 as the standard formation model S2, simulate the geological signal Gs_bound2 at the interface, and select the normalization coefficient K4;

[0022] s4.7 Establish a Picasso chart based on the instrument, predetermine the limit detection distance L for the current resistivity and contrast formation model, and forward simulate the geological signals Gs_L1 and Gs_L2 of the formation models S1 and S2 at the distance L from the instrument to the formation interface; select the initial value of the scaling factor A0, perform common logarithmic transformation on abs(Gs_L1) and abs(Gs_L2) to obtain lg(abs(Gs_L1)) and lg(abs(Gs_L2)) as exponents, and use A0 as the base to calculate A0 lg(abs(Gs_L1)) 、A0 lg(abs(Gs_L2)) ;

[0023] s4.8 Using the inversion algorithm to obtain the optimal scaling factor A, the objective functions of the formation models S1 and S2 are , , where p is the minimum proportional coefficient between geological signal and apparent resistivity signal; the geological signal is scaled and processed with the optimal scaling factors A1 and A2 as the base and the orientation is extracted to obtain , .

[0024] Optionally, in step s5, the step of establishing a formation model integrated multi-initial value strategy specifically includes:

[0025] s5.1 Based on the LWD azimuth electromagnetic wave logging response signal library constructed in step s1, normalize the formation parameters therein, perform scaling and fidelity processing on the logging response, build a training model based on a neural network, design the network architecture and define the layer structure, determine the input and output layers of the model, build and compile the model, set the training parameters and fit the model; use the processed formation parameters and logging response as the input and output of the neural network respectively, train the model, and obtain an initial model;

[0026] s5.2 Input the logging response at any y point to determine the average accuracy of the model prediction inversion initial value Is it greater than the threshold TH? If it is greater than TH, increase the amount of data and the number of iterations, and update the weights. Otherwise, use the trained neural network model to predict the initial value of the inversion. Adjust the ratio of the apparent resistivity signal to the geological signal in the well logging response at the input depth point, ρ1 and ρ2, generate N groups of well logging response combinations and input them into the neural network model to predict N groups of initial values ​​of the inversion parameters respectively.

[0027] s5.3 Integrate instrument and geological constraints to determine the range of parameters to be inverted and the detection range of the instrument-interface distance DTB min ~DTB max , instrument layer resistivity range Rt min ~Rt max , formation contrast range CR min ~CR max , considering the parameter distribution characteristics within the constraint range, randomly generate multiple initial values ​​DTB1~DTB M 、Rt1~Rt M ,CR1~CR M and combine;

[0028] s5.4 Determine whether the formation model is a dual-interface formation model, if so, proceed to step s5.5, otherwise proceed to step s5.6;

[0029] s5.5 For the dual-interface formation model, the LWD azimuthal electromagnetic wave logging response after scaling and fidelity processing is input, and the initial value prediction of the neural network and the multi-initial value prediction under comprehensive constraints are used to perform inversion to obtain the inversion result;

[0030] s5.6 For the single-interface stratigraphic model, the absolute value of the initial value of the geological signal is taken and then a gradient inversion is performed. The absolute value of the reconstructed geological signal of the optimal inversion result is taken and then inverted again. The error sizes of the two are compared, and the result with the lower error is selected as the actual inversion result.

[0031] The beneficial effect of the present invention is that the present invention creates a while-drilling azimuth electromagnetic wave logging response signal library. Considering that the data sizes of different types of signals are extremely different, when the formation resistivity is large, the contribution of geological signals in the inversion process is very small, and the formation interface cannot be effectively identified. Therefore, a multi-mode scaling factor standard is established, and cross-mode matching scaling factors are selected in different formations to perform scaling and fidelity processing on the logging data, thereby improving the anti-interference ability of the logging data, ensuring that different types of signals contribute equally in the inversion process, and improving the inversion formation interface identification ability; the present invention establishes an integrated prediction method, randomly generates multiple initial values ​​under comprehensive constraints, combines neural network training to predict initial values, and reduces the local convergence problem caused by the selection of initial values ​​in iteration. For the single-interface formation model, the absolute value of the initial value of the geological signal is taken and then inverted, and then the error size of the positive and negative cases of the geological signal is compared, which can filter out half of the initial values, improve the inversion speed, and solve the data fidelity problem and the inversion initial value problem in the while-drilling process. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the real-time fidelity inversion method of azimuthal electromagnetic wave logging data while drilling in the present invention;

[0033] Figure 2 It is a schematic diagram of a single interface formation model in the present invention;

[0034] Figure 3A It is a schematic diagram of the sensitivity of the response signal to the tilt angle in the present invention;

[0035] Figure 3B It is a schematic diagram of the sensitivity of the response signal to the formation resistivity in the present invention;

[0036] Figure 3C It is a schematic diagram of the sensitivity of the response signal to the position of the formation interface in the present invention;

[0037] Figure 4 This is a flowchart of scaling and fidelity processing in the present invention;

[0038] Figure 5 It is a Picasso diagram of the edge detection capability of the instrument in the present invention;

[0039] Figure 6 This is a schematic diagram of a neural network training model in the present invention;

[0040] Fig. 7A It is a phase difference apparent resistivity response diagram of shallow detection of single interface formation model while drilling azimuth electromagnetic wave logging of the present invention after scaling and fidelity preservation;

[0041] Figure 7B It is a phase difference apparent resistivity response diagram of the single interface formation model of the scaled and fidelity-preserved data in the present invention, and the azimuth electromagnetic wave logging deep detection while drilling;

[0042] Figure 7C It is a geological signal response diagram of electromagnetic wave logging while drilling in the single interface formation model after scaling and fidelity preservation in the present invention;

[0043] Fig. 8A This is the inversion result diagram of the traditional method single interface formation model while drilling azimuthal electromagnetic wave logging;

[0044] Figure 8B This is a diagram of the inversion results of azimuthal electromagnetic logging while drilling using the single-interface formation model of the medium-fidelity method of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0046] Real-time inversion method for azimuthal electromagnetic logging data while drilling, such as Figure 1 As shown, the following steps are included:

[0047] s1. Considering the different resistivity, contrast, anisotropy, well inclination and relative position of the instrument in the formation structure, the vertical resistivity of the target formation is set to R1=1 Ω·m, R2=10 Ω·m, R3=50 Ω·m, R4=200 Ω·m, R5=500 Ω·m, the formation contrast (instrument layer resistivity / surrounding rock layer resistivity) is CR1=2, CR2=3, CR3=5, CR4=8, CR5=10, the anisotropy coefficient is λ1=1, λ2=2, λ3=5, λ4=8, λ5=10, the well inclination angle is θ1=0°, θ2=30°, θ3=45°, θ4=60°, θ5=90°, etc. 625 formation models are obtained by combining R1~R5, CR1~CR5, λ1~λ5, θ1~θ5 respectively. Figure 2 The figure is a schematic diagram of one of the formation models. Assuming the instrument interface depth is 0 m, the instrument passes through the formation from the vertical depth of -5 m to 5 m from top to bottom, and a forward simulation is performed every 0.05 m. All simulation responses are combined to obtain the while-drilling azimuthal electromagnetic wave logging response signal library.

[0048] s2. Based on the response characteristics of apparent resistivity and geological signals in the response signal library, analyze the sensitivity of the response signal to different geological parameters, perform equivalent processing on the complex three-dimensional model, and design the LWD azimuthal electromagnetic wave logging inversion model.

[0049] According to the response characteristics of phase difference apparent resistivity Rp and geological signal Gs in the response signal library, the sensitivity of the response signal to resistivity, contrast, anisotropy, well deviation and relative position S is analyzed. M Strong or weak, design a simplified while-drilling azimuth electromagnetic wave logging inversion model; combined with Figure 3A , Figure 3B , Figure 3C The sensitivity analysis results shown in the figure show that the response signal is most sensitive to the position of the formation interface, so the inversion of the instrument-interface distance is given priority, followed by the formation resistivity, and then the well deviation. In this embodiment, the number of responses is three, and the inversion model is simplified to a single interface inversion model. The inversion parameters include: instrument layer formation resistivity Rt, surrounding rock layer formation resistivity Rs, and instrument-formation interface distance DTB.

[0050] s3. For typical formation models with different formation resistivity and contrast, a scaling factor function is constructed to obtain the best match between apparent resistivity and geological signals and optimize the objective function.

[0051] The objective function is mainly composed of the fitting difference between the measured data and the simulated data. A regularization term is added to the regularized gradient algorithm, as shown in formula (1), where e(x) is the fitting difference between the measured data and the simulated data, the second term is the regularization term, χ is the regularization parameter, x is the inversion model, and x p is the prior information; the purpose of inversion is to minimize the objective function. The convergence and smoothness of the objective functions corresponding to different forms of measured data and simulated data are very different, and there is a gap in the corresponding inversion efficiency.

[0052] (1)

[0053] s4. Eliminate the base value of the amplitude ratio signal, perform logarithmic processing on the amplitude ratio and phase difference signals, determine the normalization coefficients of the amplitude ratio, phase difference and geological signal, select the optimal scaling factor of the geological signal, and automatically process the data using the scaling fidelity method; the steps of the scaling fidelity processing method are as follows: Figure 4 As shown, specifically including:

[0054] s4.1 Considering that the instrument response signal only has phase difference resistivity signal, the logarithm of the two phase difference signals after reverse fitting is taken to obtain log 10 (Ps1) and log 10 (Ps2), select log10 (Ps1) and log 10 Normalization coefficients K1 and K2 of (Ps2) are calculated and The two apparent resistivity signals are normalized respectively;

[0055] s4.2 For the geological signal, set the original value of FLAG to -1, and determine whether the geological signal Gs is greater than zero. If it is greater than zero, assign the FLAG value to 1, otherwise execute step s4.3;

[0056] s4.3 Perform an absolute value operation on the Gs signal to obtain abs(Gs), and determine whether the abs(Gs) signal is less than the signal threshold. If so, assign the abs(Gs) signal to the signal threshold. Otherwise, execute step s4.4.

[0057] s4.4 Determine whether there is prior information of adjacent well data. If so, perform a forward simulation to calculate the geological signal Gs_bound1 at the interface based on the formation model S1 of the current layer resistivity Rt1 and the contrast CR1 in the prior information, and use Gs_bound1 to select the normalization coefficient K3, otherwise execute step s4.5;

[0058] s4.5 Based on the shallow detection apparent resistivity signal Ra as the instrument layer resistivity Rt2, with the formation contrast CR2 (ratio of surrounding rock resistivity to instrument layer resistivity) = 1:5 as the standard formation model S2, simulate the geological signal Gs_bound2 at the interface, and select the normalization coefficient K4;

[0059] s4.6 As Figure 5 As shown in the figure, a Picasso chart is established based on the instrument, and the limit detection distance L is pre-determined for the formation model of the current resistivity and contrast. The geological signals Gs_L1 and Gs_L2 of the formation models S1 and S2 at the distance L from the instrument to the formation interface are simulated forward. The initial value of the scaling factor A0 is selected, and the common logarithmic transformation of abs(Gs_L1) and abs(Gs_L2) is performed to obtain lg(abs(Gs_L1)) and lg(abs(Gs_L2)) as exponents, and A0 is calculated with A0 as the base. lg(abs(Gs_L1)) 、A0 lg(abs(Gs_L2)) ;

[0060] s4.7 Using the inversion algorithm to obtain the optimal scaling factor A, the objective functions of the formation models S1 and S2 are , , where p is the minimum proportional coefficient between geological signal and apparent resistivity signal; the geological signal is scaled and processed with the optimal scaling factors A1 and A2 as the base and the orientation is extracted to obtain , .

[0061] s5. Introduce a neural network algorithm, combine the shallow detection apparent resistivity value and prior information constraints, establish a formation model integrated multi-initial value strategy, and optimize the initial value based on the actual logging data; the steps of establishing the formation model integrated multi-initial value strategy specifically include:

[0062] s5.1 Based on the LWD azimuth electromagnetic wave logging response signal library constructed in step s1, normalize the formation parameters therein and perform scaling and fidelity processing on the logging response. Figure 6 As shown, a training model based on a neural network is constructed, the network architecture is designed and the layer structure is defined, the input and output layers of the model are determined, the model is built and compiled, the training parameters are set and the model is fitted; the processed formation parameters and logging response are used as the input and output of the neural network respectively, the model is trained to obtain an initial model;

[0063] s5.2 Input the logging response at any y point to determine the average accuracy of the model prediction inversion initial value Is it greater than the threshold TH? If it is greater than TH, increase the amount of data and the number of iterations, and update the weights. Otherwise, use the trained neural network model to predict the initial value of the inversion. Adjust the ratio of the apparent resistivity signal to the geological signal in the well logging response at the input depth point, ρ1 and ρ2, generate N groups of well logging response combinations and input them into the neural network model to predict N groups of initial values ​​of the inversion parameters respectively.

[0064] s5.3 Comprehensively consider the instrument and geological constraints to determine the range of parameters to be inverted. The detection range of the instrument-interface distance DTB is 0-5 m, the instrument layer resistivity range is 1-50 Ω·m, and the formation contrast range is 1-10. Considering the parameter distribution characteristics within the constraint range, multiple initial values ​​are randomly generated: the initial values ​​of the instrument-interface distance are 0.5 m, 2 m, and 4 m; the initial values ​​of the instrument layer resistivity are , Rp, (Rp is the shallow detection phase difference apparent resistivity); the initial values ​​of formation contrast are 2, 5, and 8 respectively; they are combined to form 27 initial formation models;

[0065] s5.4 For the inversion of the single-interface stratigraphic model DTB, the absolute value of the initial value of the geological signal is taken for inversion. First, the absolute distance between the instrument and the interface is determined. Then, the inverse of the reconstructed geological signal of the inversion result is taken and re-inverted. The error of the inversion results before and after is compared. The inversion results before and after are the cases where the interface is located at the same distance above and below the instrument. Finally, the low-error result is selected as the actual inversion result, eliminating the need for the positivity of the initial value of DTB in the upper and lower directions, reducing the initial value of DTB by half.

[0066] s6. Input Fig. 7A , Figure 7Band Figure 7C The azimuthal electromagnetic logging data while drilling is shown. Fig. 7A The solid line and dashed line in the middle are the measured phase difference curve of shallow detection after scaling and the reconstructed curve based on the simulation of inversion results. Figure 7B The solid line and dashed line in the middle are the measured curve of deep detection phase difference after scaling and the reconstructed curve based on the simulation of inversion results. Figure 7C The solid line and dashed line in the middle are the measured curve of geological signals after scaling and the reconstructed curve based on the simulation of inversion results. 0 ~10 2 , the maximum value of geological signal is less than 10 -2 , the contribution of the two signals in the objective function is significantly different, and the "horn" of the apparent resistivity signal near the formation interface will seriously suppress the geological signal; in the data after scaling and fidelity preservation, the two signals are stable at the same order of magnitude, and the apparent resistivity signal is converted into a phase difference to avoid the influence of the "horn" on the geological signal near the interface. In addition, the geological signal after processing is smoother than the original data, which increases the depth range of the effective contribution of the geological signal; based on the regularized gradient optimization algorithm and the initial value of step s5, the upper / lower interface position and resistivity distribution of the instrument are extracted in real time.

[0067] s7. Using traditional methods to invert, we get Fig. 8A The inversion results shown in the figure are inverted using the fidelity method and statistically processed for multiple groups of inversion results to obtain Figure 8B The final inversion results shown in the figure finally obtain the final formation model and the credibility of each inversion parameter. The comparison between the inversion results of the traditional method and the inversion results of the fidelity method shows that after the fidelity processing, the instrument has a stronger ability to identify the formation interface and can more effectively protect the authenticity of the formation information in the logging data.

[0068] Experiments have shown that the inverted resistivity is highly consistent with the formation model value, verifying the accuracy and reliability of the inversion algorithm.

[0069] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A real-time inversion method for azimuthal electromagnetic logging data while drilling, characterized in that: The steps include: s1. Establish various horizontal formation models, carry out forward simulation of azimuthal electromagnetic logging while drilling, and obtain the azimuthal electromagnetic logging while drilling response signal library; s2. Based on the response signal library, analyze the sensitivity of the response signal to different geological parameters and design the inversion model of azimuthal electromagnetic wave logging while drilling; s3. Input a apparent resistivity curve and b geological signal curve, establish the reverse fitting relationship between apparent resistivity and amplitude ratio and phase difference signal, and convert the apparent resistivity curve into amplitude ratio and phase difference signal; s4. Eliminate the base value of the amplitude ratio signal, perform logarithmic processing on the amplitude ratio and phase difference signals, determine the normalization coefficients of the amplitude ratio, phase difference and geological signal, select the optimal scaling factor of the geological signal, and automatically process the data using the scaling fidelity method; s5. Introduce neural network algorithm, establish formation model integrated multi-initial value strategy, and optimize initial value based on actual logging data; s6. Based on the regularized gradient optimization algorithm and the initial value of step s5, the upper / lower interface position and resistivity distribution of the instrument are extracted in real time; s7. Perform statistical processing on multiple sets of inversion results to obtain the final formation model and the credibility of each inversion parameter.

2. The method for real-time inversion of azimuthal electromagnetic logging data while drilling according to claim 1, characterized in that: In step s4, the steps of the scaling fidelity processing method specifically include: s4.1 For the amplitude ratio signal Att, simulate the amplitude base value A0 in the air medium, calculate Att_p=(Att-Att0) as the net amplitude ratio signal, and take the logarithm to get log 10 (Att_p), similarly, take the logarithm of the phase difference signal to get log 10 (Ps); s4.2 Select log 10 The normalization coefficient K1 of (Att_p) is calculated Get normalized, select log 10 The normalization coefficient K2 of (Ps) is calculated Get normalized; s4.3 For the geological signal, set the original value of FLAG to -1, and determine whether the geological signal Gs is greater than zero. If it is greater than zero, assign the FLAG value to 1, otherwise execute step s4.4; s4.4 Perform an absolute value operation on the Gs signal to obtain abs(Gs), and determine whether the abs(Gs) signal is less than the signal threshold. If so, assign the abs(Gs) signal to the signal threshold. Otherwise, execute step s4.

5. s4.5 Determine whether there is prior information of adjacent well data. If so, perform a forward simulation to calculate the geological signal Gs_bound1 at the interface based on the formation model S1 of the current layer resistivity Rt1 and the contrast CR1 in the prior information, and use Gs_bound1 to select the normalization coefficient K3, otherwise execute step s4.6; s4.6 According to the shallow detection apparent resistivity signal Ra as the instrument layer resistivity Rt2, the formation contrast CR2 = 1:5 is used as the standard formation model S2, the geological signal Gs_bound2 at the interface is simulated, and the normalization coefficient K4 is selected; s4.7 Establish a Picasso chart based on the instrument, predetermine the limit detection distance L for the current resistivity and contrast formation model, and forward simulate the geological signals Gs_L1 and Gs_L2 of the formation models S1 and S2 at the distance L from the instrument to the formation interface; select the initial value of the scaling factor A0, perform common logarithmic transformation on abs(Gs_L1) and abs(Gs_L2) to obtain lg(abs(Gs_L1)) and lg(abs(Gs_L2)) as exponents, and use A0 as the base to calculate A0 lg(abs(Gs_L1)) 、A0 lg(abs(Gs_L2)) ; s4.8 Using the inversion algorithm to obtain the optimal scaling factor A, the objective functions of the formation models S1 and S2 are , , where p is the minimum proportional coefficient between geological signal and apparent resistivity signal; the geological signal is scaled and processed with the optimal scaling factors A1 and A2 as the base and the orientation is extracted to obtain , .

3. The method for real-time inversion of azimuthal electromagnetic logging while drilling data according to claim 2, characterized in that: In step s5, the steps of establishing a formation model integrated multi-initial value strategy specifically include: s5.1 Based on the LWD azimuth electromagnetic wave logging response signal library constructed in step s1, normalize the formation parameters therein, perform scaling and fidelity processing on the logging response, build a training model based on a neural network, design the network architecture and define the layer structure, determine the input and output layers of the model, build and compile the model, set the training parameters and fit the model; use the processed formation parameters and logging response as the input and output of the neural network respectively, train the model, and obtain an initial model; s5.2 Input the logging response at any y point to determine the average accuracy of the model prediction inversion initial value Is it greater than the threshold TH? If it is greater than TH, increase the amount of data and the number of iterations, and update the weights. Otherwise, use the trained neural network model to predict the initial value of the inversion. Adjust the ratio of the apparent resistivity signal to the geological signal in the well logging response at the input depth point, ρ1 and ρ2, generate N groups of well logging response combinations and input them into the neural network model to predict N groups of initial values ​​of the inversion parameters respectively. s5.3 Integrate instrument and geological constraints to determine the range of parameters to be inverted and the detection range of the instrument-interface distance DTB min ~DTB max , instrument layer resistivity range Rt min ~Rt max , formation contrast range CR min ~CR max , considering the parameter distribution characteristics within the constraint range, randomly generate multiple initial values ​​DTB1~DTB M 、Rt1~Rt M ,CR1~CR M and combine; s5.4 Determine whether the formation model is a dual-interface formation model, if so, proceed to step s5.5, otherwise proceed to step s5.6; s5.5 For the dual-interface formation model, the LWD azimuthal electromagnetic wave logging response after scaling and fidelity processing is input, and the initial value prediction of the neural network and the multi-initial value prediction under comprehensive constraints are used to perform inversion to obtain the inversion result; s5.6 For the single-interface stratigraphic model, the absolute value of the initial value of the geological signal is taken and then a gradient inversion is performed. The absolute value of the reconstructed geological signal of the optimal inversion result is taken and then inverted again. The error sizes of the two are compared, and the result with the lower error is selected as the actual inversion result.

Images