Feature extraction method for downhole transient electromagnetic signals

Abstract

The application provides a feature extraction method of a downhole transient electromagnetic signal, and the method comprises the following steps: obtaining a preset feature extraction model; obtaining measured transient electromagnetic response signals at different well depths in a well to be measured, and obtaining target feature parameters at the corresponding well depths according to the measured transient electromagnetic response signals, wherein the target feature parameters are feature parameters with a correlation degree greater than a preset correlation degree with the water invasion degree at different well depths; and substituting the target feature parameters into the preset feature extraction model to obtain the topographic features at the corresponding well depths in the well to be measured. The feature extraction method of the downhole transient electromagnetic signal provided by the application makes the extracted features more comprehensive, and thus can effectively identify the formation-related information at different well depths, so that the evaluation of the remaining oil and gas in the well is also more accurate.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field logging technology, and in particular to a method for feature extraction of downhole transient electromagnetic signals. Background Technology

[0002] In oil and gas field development, the status and distribution of remaining oil and gas in underground reservoirs are crucial objectives for oilfield development research and the foundation for major development decisions. Determining the status and distribution of remaining oil and gas requires timely and accurate monitoring and evaluation. Existing technology proposes using transient electromagnetic induction logging (TEMP) for monitoring remaining oil and gas. TEMP measures the secondary induced electromotive force generated in the formation; however, it is easily affected by factors such as casing electrical parameters, cement sheaths, and invasion zones, leading to discrepancies between the measured transient electromagnetic response signal and the actual formation information. To accurately calculate formation resistivity, effective feature extraction of the acquired transient electromagnetic signals is essential.

[0003] Currently, the methods for feature extraction of transient electromagnetic signals are relatively simple and the extracted features are not comprehensive, which makes the obtained formation-related information and residual oil and gas assessments not very accurate. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed to provide a feature extraction method for downhole transient electromagnetic signals that overcomes or at least partially solves the above problems. It can solve the problems that existing transient electromagnetic feature extraction methods are too simplistic and the extracted features are not comprehensive, resulting in inaccurate formation-related information and residual oil and gas evaluation, and achieve the technical effect of better distinguishing the characteristics of residual oil and gas in the formation.

[0005] Specifically, the present invention provides a feature extraction method for downhole transient electromagnetic signals, characterized by comprising:

[0006] Obtain the preset feature extraction model;

[0007] The measured transient electromagnetic response signals at different depths in the well to be tested are obtained, and the target feature parameters at the corresponding depths are obtained based on the measured transient electromagnetic response signals. The target feature parameters are feature parameters whose correlation with the degree of water intrusion at different depths is greater than a preset correlation.

[0008] The target feature parameters are substituted into the preset feature extraction model to obtain the terrain features at the corresponding well depth of the well to be measured.

[0009] Optionally, the method for obtaining the preset feature extraction model includes:

[0010] Acquire standard transient electromagnetic response signals at different depths in a standard well;

[0011] Calculate the characteristic parameters of the standard transient electromagnetic response signal in multiple dimensions;

[0012] Obtain the correlation between each of the aforementioned feature parameters and the degree of water intrusion at different well depths, and select the feature parameters whose correlation with the degree of water intrusion is greater than the preset correlation as target feature parameters;

[0013] The preset feature extraction model is established based on the target feature parameters.

[0014] Optionally, before calculating the feature parameters of the standard transient electromagnetic response signal in multiple dimensions, the method for obtaining the preset feature extraction model further includes:

[0015] The standard transient electromagnetic response signal is preprocessed.

[0016] Optionally, the data preprocessing of the standard transient electromagnetic response signal includes: data deduplication, missing value processing, outlier processing, and / or data noise reduction.

[0017] Optionally, the multiple dimensions include statistical dimensions, time-frequency domain dimensions, shape metric dimensions, and / or hysteresis dimensions.

[0018] Optionally, the feature values ​​of the statistical dimension include maximum value, minimum value, mean, variance, standard deviation, peak value, kurtosis and / or skewness;

[0019] The eigenvalues ​​of the time-frequency domain dimension include the centroid frequency, average frequency, root mean square frequency, and / or frequency variance.

[0020] The eigenvalues ​​of the shape metric dimension include Shannon entropy, Higitch fractal dimension, Katz fractal dimension, and / or generalized Hearst exponent;

[0021] The eigenvalues ​​of the lag dimension include the autocorrelation eigenvalues ​​of the time series.

[0022] Optionally, obtaining the correlation between each of the characteristic parameters and the degree of water intrusion at different well depths includes:

[0023] A preset correlation analysis algorithm is used to calculate the correlation between the characteristic parameters and the degree of water intrusion at different well depths.

[0024] Optionally, establishing the preset feature extraction model based on the target feature parameters includes:

[0025] The preset feature model is trained by using the target feature parameters as input and the water intrusion degree corresponding to different well depths as output.

[0026] Weights are assigned to the target feature parameters to establish the preset feature extraction model.

[0027] Optionally, assigning weights to the target feature parameters includes:

[0028] An adaptive enhancement ensemble algorithm is used to assign weights to the target feature parameters.

[0029] Optionally, the step of establishing the preset feature extraction model based on the target feature parameters further includes:

[0030] A regression random forest algorithm was used to obtain a nonlinear relationship data model between the standard transient electromagnetic response signal and the degree of water intrusion at different well depths by fitting a classification regression tree.

[0031] The nonlinear relational data model is used as the preset feature extraction model.

[0032] In the feature extraction method for downhole transient electromagnetic signals of the present invention, a preset feature extraction model is first obtained. This preset feature extraction model can be a pre-established standard feature extraction model, which is obtained through multiple data calculations and can effectively obtain corresponding detailed feature information by substituting parameters. Then, a transient electromagnetic logging instrument is lowered into the well to be logged, and the measured transient electromagnetic response signals at different depths in the well are measured. Next, the obtained measured transient electromagnetic response signals are further optimized, and feature parameters with a correlation greater than a preset correlation with the degree of water intrusion at different depths in the well are selected, and these feature parameters are used as target feature parameters. Finally, the above target feature parameters are substituted into the preset feature extraction model to obtain the topographic features of the formation with the degree of water intrusion at different depths corresponding to the measured transient electromagnetic signals. By incorporating a preset feature extraction model and filtering the feature parameters of the measured transient electromagnetic response signals at different well depths, the extraction of formation feature information corresponding to the measured electromagnetic response signals in the well to be logged is more comprehensive and accurate. This, in turn, makes the relevant information on the degree of water invasion in different formations and the evaluation of remaining oil and gas more accurate.

[0033] The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description

[0034] The following sections will describe some specific embodiments of the invention in detail by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or portions. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0035] Figure 1 This is a schematic flowchart of a feature extraction method for downhole transient electromagnetic signals according to an embodiment of the present invention;

[0036] Figure 2 This is a schematic flowchart of a feature extraction method according to another embodiment of the present invention;

[0037] Figure 3 This is a schematic flowchart of a feature extraction method according to another embodiment of the present invention;

[0038] Figure 4 This is a schematic flowchart of a feature extraction method according to another embodiment of the present invention;

[0039] Figure 5 This is a signal feature map of a downhole transient electromagnetic signal before data preprocessing according to an embodiment of the present invention;

[0040] Figure 6 This is a signal feature map after data preprocessing of downhole transient electromagnetic signals according to an embodiment of the present invention. Detailed Implementation

[0041] The following reference Figures 1 to 6 This invention describes a method for feature extraction of downhole transient electromagnetic signals according to embodiments of the present invention. In this description, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" and "second" may explicitly or implicitly include at least one of that feature, that is, include one or more of that feature. In the description of the present invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. When a feature "includes or contains" one or more of the features it encompasses, unless otherwise specifically described, this indicates that other features are not excluded and may be further included.

[0042] In the description of this embodiment, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0043] Figure 1This is a schematic flowchart of a feature extraction method for downhole transient electromagnetic signals according to an embodiment of the present invention, as shown below. Figure 1 As shown, and refer to Figures 2 to 6 This invention provides a method for feature extraction of downhole transient electromagnetic signals, which includes the following steps:

[0044] Step S1: Obtain the preset feature extraction model;

[0045] Step S2: Obtain the measured transient electromagnetic response signals at different depths in the well to be tested, and obtain the target feature parameters at the corresponding depths based on each measured transient electromagnetic response signal. The target feature parameters are those with a correlation greater than a preset correlation with the degree of water intrusion at different depths in the well.

[0046] Step S3: Substitute the target feature parameters into the preset feature extraction model to obtain the terrain features at the corresponding well depth of the well to be measured.

[0047] In step S1 above, a preset feature extraction model is obtained. This preset feature extraction model can be a pre-established standard feature extraction model, which is obtained through multiple data calculations. Detailed feature information can be obtained by substituting parameters into the feature extraction model.

[0048] In step S2 above, a transient electromagnetic logging instrument is inserted into the well to be logged, and the measured transient electromagnetic response signals at different depths in the well are measured. Then, the obtained measured transient electromagnetic response signals are further optimized, and feature parameters with a correlation greater than a preset correlation with the degree of water invasion at different depths in the well are selected, and these feature parameters are used as target feature parameters.

[0049] In step S3 above, the target feature parameters are substituted into the preset feature extraction model to obtain the topographic features of the formation with water intrusion degree at different well depths corresponding to the measured transient electromagnetic signals.

[0050] The feature extraction method provided in this embodiment, by pre-setting a feature extraction model and screening the feature parameters of the measured transient electromagnetic response signals at different well depths, makes the extraction of formation feature information corresponding to the measured electromagnetic response signals in the well to be logged more comprehensive and accurate. In this way, it can effectively identify formation-related information at different well depths, and make the evaluation of the remaining oil and gas in the well more accurate.

[0051] In some embodiments of the present invention, such as Figure 2 As shown, the method for obtaining the preset feature extraction model in step S1 above includes:

[0052] Step S101: Obtain standard transient electromagnetic response signals at different depths in a standard well;

[0053] Step S103: Calculate the characteristic parameters of the standard transient electromagnetic response signal in multiple dimensions;

[0054] Step S104: Obtain the correlation between each feature parameter and the degree of water invasion at different well depths, and select the feature parameters whose correlation with the degree of water invasion is greater than the preset correlation as target feature parameters;

[0055] Step S105: Establish a preset feature extraction model based on the target feature parameters.

[0056] This embodiment presents a method for establishing a preset feature extraction model. First, a transient electromagnetic logging instrument is deployed into a standard well to measure the standard transient electromagnetic response signals at different depths. Then, multiple dimensions of feature parameters for each standard transient electromagnetic response signal are calculated, and feature parameters with a correlation greater than a preset correlation value to the degree of water intrusion are selected as target feature parameters. Finally, a preset feature extraction model is established based on the target feature parameters. The preset correlation value can be set as the correlation coefficient of the feature parameters with an absolute value of 0.5.

[0057] By establishing a preset feature extraction model using the above method, the data dimensionality of the collected measured transient electromagnetic response signals can be reduced, the data sample quality can be improved, and the efficiency of establishing the preset feature extraction model can be increased. At the same time, the accuracy of feature extraction through the preset feature extraction model can also be improved.

[0058] In some embodiments of the present invention, such as Figure 3 As shown, before calculating the feature parameters of the standard transient electromagnetic response signal in multiple dimensions, the method for obtaining the preset feature extraction model also includes:

[0059] S102 performs data preprocessing on the standard transient electromagnetic response signal.

[0060] In this embodiment, as Figure 5 and Figure 6 As shown, taking the measured downhole transient electromagnetic response signal at a depth of 2230m-2270m as an example, data preprocessing methods can be used to obtain signal feature maps before and after preprocessing. By preprocessing the standard transient electromagnetic response signal, the data samples can be optimized, making the processed electromagnetic response signal samples purer and facilitating subsequent calculation of the multi-dimensional feature parameters of the electromagnetic response signal. It is worth noting that h in this embodiment... j The well depth is 2230m-2270m.

[0061] In some embodiments of the present invention, data preprocessing of the standard transient electromagnetic response signal includes: data deduplication.

[0062] In this embodiment, when performing data preprocessing on the electromagnetic response signal, data deduplication can be performed on the standard transient electromagnetic response signal, mainly by removing duplicate measured data. This effectively avoids repeated calculations of the standard transient electromagnetic response signal, reduces its storage space occupation, and improves data processing efficiency.

[0063] Furthermore, in order to optimize the sample data, missing value processing can also be performed on the data samples of the standard transient electromagnetic response signal. In the process of missing value processing, direct deletion, missing value filling, or missing value fitting can be used to improve the data samples and enhance the accuracy and reliability of the sample data.

[0064] Furthermore, outlier processing can also be performed on the data samples of the standard transient electromagnetic response signal. In the process of outlier processing, methods such as deleting outliers, replacing outliers, binning, using outliers for feature engineering, or building robust models can be used to further improve the accuracy and reliability of the sample data.

[0065] Furthermore, data denoising can also be performed simultaneously on the standard transient electromagnetic response signal data samples. During the data denoising process, methods such as mean filtering, median filtering, wavelet transform, connected graph methods, or support vector machines can be used to further improve the accuracy and reliability of the sample data. Similarly, zero-drift baseline correction and recursive filtering methods can also be used for signal denoising.

[0066] Of course, in order to achieve better data preprocessing results, data deduplication, missing value processing, outlier processing, and data noise reduction can be performed simultaneously on the data samples of the standard transient electromagnetic response signal.

[0067] Specifically, in the actual data preprocessing operation, the depth h of the measured well depth is used as the basis for the data preprocessing. j Using a unique identifier, repetitive transient electromagnetic response signals in continuous sampling mode are removed. A sliding window of size s=3 is used, and sliding window interpolation is employed for each depth h. j Missing values ​​were filled into the standard transient electromagnetic response signal. Outliers were identified by plotting a scatter plot of the standard transient electromagnetic response signal, using h... j-1 and h j+1The average signal value at the same time point is used to fill in the missing values, and median smoothing filtering is used to remove interference from high-frequency electromagnetic signals. The aforementioned sliding window of size s=3 refers to using the length of three transient electromagnetic response signals as the detection target during a single slide. If a transient electromagnetic response signal is missing, it can be identified more quickly, and interpolation is used to fill in the missing values. This setting improves the efficiency of missing value identification during data preprocessing and also facilitates the filling of missing values ​​later. Furthermore, median smoothing filtering is used to filter high-frequency electromagnetic signals. Median smoothing filtering mainly identifies high-frequency outliers by calculating the median after identifying the frequencies of the three transient electromagnetic response signals, and these identified high-frequency outliers are directly filtered out.

[0068] In some embodiments of the present invention, multiple dimensions include statistical dimensions, time-frequency domain dimensions, shape measurement dimensions, and / or hysteresis dimensions.

[0069] In this embodiment, each depth h j The transient electromagnetic response signal after data preprocessing is defined as the numerical value X = {x1, x2, ..., x...} i , ..., x n}

[0070] Among them, h j This represents the well depth at height j. i This represents the i-th transient electromagnetic response signal.

[0071] The calculated feature parameter values ​​of different dimensions are defined as F = {F1, F2, ..., F...} m}

[0072] Where =1 represents the multi-dimensional characteristic parameter value of the transient electromagnetic response signal corresponding to x1. m represents the number of samples of the transient electromagnetic response signal.

[0073] By calculating the feature parameters of the standard transient electromagnetic response signal in multiple dimensions and fusing the features of different dimensions, we can obtain more discriminative and comprehensive spatiotemporal features of the standard transient electromagnetic response signal. This makes the target feature parameters used to build the preset feature extraction model more comprehensive, and the established model more reliable and accurate.

[0074] In some embodiments of the present invention, the feature values ​​of the statistical dimension include maximum, minimum, mean, variance, standard deviation, peak value, kurtosis, and / or skewness. The feature values ​​of the time-frequency domain dimension include centroid frequency, average frequency, root mean square frequency, and / or frequency variance. The feature values ​​of the shape metric dimension include Shannon entropy, Higuch fractal dimension, Katz fractal dimension, and / or the generalized Hearst exponent. The feature values ​​of the lag dimension include the autocorrelation features of the time series.

[0075] In this embodiment, the statistical dimension features include the maximum value F1, minimum value F2, mean F3, variance F4, standard deviation F5, peak value F6, kurtosis F7, and skewness F8. The formulas for calculating each feature are as follows:

[0076] F1 = max(X(i));

[0077] F2 = min(X(i));

[0078]

[0079]

[0080]

[0081] F6 = F1 - F2;

[0082]

[0083]

[0084] Where X = {x1, x2, ..., x} i , ..., x n} represents each depth h j Transient electromagnetic response signal after data preprocessing; F = {F1, F2, ..., F m} represents multiple multidimensional feature parameter values ​​of different dimensions; n represents the length of the transient electromagnetic response signal X; m represents the number of samples in the transient electromagnetic response signal.

[0085] The eigenvalues ​​in the time-frequency domain include the calculated centroid frequency F9 and the average frequency F. 10 , frequency root mean square F 11 Frequency variance F 12 The formulas for calculating each eigenvalue are as follows:

[0086]

[0087]

[0088]

[0089]

[0090] In the above formula, k represents the number of intervals after segmenting the transient electromagnetic response signal X; fk represents the frequency amplitude of the corresponding point of X; s(k) represents the power spectrum value of X; and m represents the number of samples in the transient electromagnetic response signal.

[0091] The eigenvalues ​​of the shape metric dimension include the calculation of Shannon entropy F. 13 Higashich fractal dimension F 14 Katz fractal dimension F 15 Generalized Hearst exponent F 16 The formulas for calculating each eigenvalue are as follows:

[0092]

[0093]

[0094]

[0095]

[0096] In the above formula, p(x) i The expression () represents the proportion of the number of variable categories to which a single sample variable belongs out of the total number of variables; in other words, it yields x. i The probability of the value;

[0097] L(k) represents the average value of Lm(k) with respect to each k, taking m values.

[0098] L represents the total length of the signal;

[0099] d represents the planar distance of the waveform;

[0100] τ represents the time delay within the time window;

[0101] q represents the order;

[0102] t = υ, 2υ, ..., kυ, T. Where T represents the observation period and υ represents the time resolution.

[0103] The eigenvalues ​​of the lag dimension include the autocorrelation eigenvalue F of the time series. 17 The calculation formula is as follows:

[0104]

[0105] Where μ represents the overall mean of the transient electromagnetic response signal X;

[0106] l represents the lag coefficient;

[0107] σ represents the standard deviation of the transient electromagnetic response signal X.

[0108] By calculating the above multi-dimensional feature parameter values, we can obtain the feature parameter values ​​of different dimensions corresponding to each transient electromagnetic response signal. This facilitates feature fusion of feature parameters of different dimensions, enabling us to obtain more discriminative and comprehensive spatiotemporal features of standard transient electromagnetic response signals. Consequently, the target feature parameters used to establish the preset feature extraction model become more comprehensive, and the established model becomes more reliable and accurate.

[0109] In some embodiments of the present invention, the correlation between each characteristic parameter and the degree of water intrusion at different well depths is obtained, including by using a preset correlation analysis algorithm to calculate the correlation between the characteristic parameters and the degree of water intrusion at different well depths.

[0110] In this embodiment, firstly, based on water intrusion data from standard wells, a preset correlation analysis algorithm is used to screen characteristic parameters that significantly influence the degree of water intrusion. The preset correlation analysis algorithm can be the Pearson correlation analysis algorithm. For each depth h... j The transient electromagnetic response signal X = {x1, x2, ..., x} i , ..., x n}, multidimensional feature parameter values ​​F = {F1, F2, ..., F} m The sequence of water intrusion values ​​is Y = {y1, y2, ..., y}. j This forms a j-row (m+1)-column two-dimensional matrix with depth points as rows and feature parameters and water intrusion levels as columns. The feature parameter F for each column is calculated. i Pearson correlation coefficient with the (n+1)th column Y sequence in

[0111] Meanwhile, the Pearson correlation coefficient mentioned above ranges from -1 to 1. A value closer to 1 or -1 indicates a stronger linear relationship between the two variables, while a value closer to 0 indicates a weaker or no linear relationship. An absolute value greater than or equal to 0.5 indicates a strong linear relationship. The calculated correlation coefficient set R = {r1, r2, ..., r...} m The algorithm filters feature parameters with an absolute value greater than or equal to 0.5 as target feature parameters to further improve the efficiency of extracting transient electromagnetic response signals at different well depths. This allows for faster screening of feature parameters that are highly correlated with the degree of water intrusion at different well depths. The preset correlation can be set to an absolute value of 0.5, with a correlation greater than the preset correlation used as the screening criterion.

[0112] For example, the Pearson correlation coefficient for F1 is 0.37, and the Pearson correlation coefficient for F17 is 0.78, etc. See the table below.

[0113]

[0114] In the table above, the Pearson values ​​represent the Pearson correlation coefficient. It is easy to see from the table that using the multi-dimensional feature parameter values ​​corresponding to an absolute Pearson correlation coefficient greater than or equal to 0.5 as the target feature parameters yields the highest accuracy in determining the degree of water intrusion at different well depths.

[0115] In other embodiments of the present invention, the correlation analysis method can also employ other correlation analysis methods, such as grey relational analysis. The basic idea of ​​grey relational analysis is to determine the closeness of the connection between different sequences based on the similarity of their geometric shapes. Its basic approach is to transform the discrete behavioral observations of system factors into piecewise continuous polygonal lines using linear interpolation, and then construct a model to measure the degree of correlation based on the geometric characteristics of the polygonal lines. The closer the geometric shapes of the polygonal lines are, the greater the correlation between the corresponding sequences, and vice versa. In practical applications, a data sequence reflecting the characteristics of system behavior can be used as a reference sequence. A data sequence composed of factors affecting system behavior can be used as a comparison sequence. The correlation degree between the comparison sequence and the reference sequence at each time point is used as the correlation degree. Each depth h can be used as the reference sequence. j The transient electromagnetic response signal is used as a comparison sequence, and the water intrusion degree value is used as a reference sequence to obtain the correlation degree between the two. Then, by comparing the magnitude of each correlation degree, the influence of the characteristic parameters of different transient electromagnetic response signals on the water intrusion degree at different well depths is determined.

[0116] In some embodiments of the present invention, establishing a preset feature extraction model based on target feature parameters includes:

[0117] The target feature parameters are used as the input to the preset feature extraction model, and the water intrusion degree at different well depths is used as the output of the preset feature extraction model to train the preset feature extraction model.

[0118] Weights are assigned to the target feature parameters to establish a pre-defined feature extraction model.

[0119] In this embodiment, to establish a preset feature extraction model, the selected feature parameters are directly used as target feature parameters. These target feature parameters are used as input, while the corresponding water intrusion levels at different well depths are used as outputs to train the preset feature extraction model. Weights are then assigned to each target feature parameter. For example, when assigning weights, feature parameters highly correlated with water intrusion levels at different well depths are assigned larger weights to obtain a more accurate preset feature extraction model.

[0120] In some embodiments of the present invention, assigning weights to the target feature parameters includes using an adaptive enhancement ensemble algorithm to assign weights to the target feature parameters.

[0121] In this embodiment, the adaptive augmentation ensemble algorithm is an adaptive boosting algorithm. The core idea of ​​the adaptive augmentation ensemble algorithm is to sequentially train a series of dependent, similar models, where each subsequent model is used to correct the output of the preceding model. After fitting a series of sequential models, a more accurate ensemble model is ultimately obtained. By assigning weights to the target feature parameters through the adaptive augmentation ensemble algorithm, the weights allocated to each feature parameter are more reasonable, resulting in higher accuracy. This achieves the goal of automatically assigning weights to each feature parameter, thus increasing efficiency.

[0122] In some embodiments of the present invention, establishing a preset feature extraction model based on target feature parameters further includes:

[0123] A regression random forest algorithm was used to obtain a nonlinear data model of the relationship between the standard transient electromagnetic response signal and the degree of water intrusion at different well depths by fitting a classification regression tree.

[0124] The nonlinear relational data model is used as the preset feature extraction model.

[0125] In this embodiment, a quantitative analysis data model based on the regression random forest method is mainly used to construct a multi-dimensional feature parameter F and the degree of water intrusion Y. During sample training, the training set consists of combined feature quantities extracted from 2100 sets of transient electromagnetic response signals and the corresponding water intrusion values ​​at different well depths. The training dataset E = {(U1,Y1),(U2,Y2),…,(U…} 2100 ,Y 2100 )}, where U i ={F i1 ,F i2 ,…F im} represents the combined characteristic quantity of transient electromagnetic signals at a certain depth, Y i This represents the water intrusion level corresponding to the well depth. After training with 2100 sets of samples, a nonlinear relationship model was established between the combined features F of transient electromagnetic signals and the water intrusion level Y, based on a regression random forest algorithm. Then, 900 sets of test samples were used to test and refine the established nonlinear relationship model, resulting in the final quantitative analysis model. This quantitative analysis model is the preset feature extraction model.

[0126] Furthermore, this embodiment primarily employs random forest regression for quantitative analysis and detection to obtain a data model relating the transient electromagnetic response signal and the degree of water intrusion. Due to the nonlinear relationship between the transient electromagnetic response signal and the degree of water intrusion, the random forest method can train a classifier containing multiple decision trees to obtain a more accurate model of the nonlinear relationship between the transient electromagnetic response signal U and the degree of water intrusion Y. This more accurately fits the nonlinear relationship between the transient electromagnetic response signal and the degree of water intrusion, thereby making the obtained preset feature extraction model more accurate and reliable.

[0127] In other embodiments of the present invention, the randomized synergistic algorithm used in the above embodiments may also be other machine learning algorithms, such as neural networks.

[0128] In some embodiments of the present invention, such as Figure 4 As shown, the feature extraction method for downhole transient electromagnetic signals also includes:

[0129] S4 verifies the effectiveness of the preset feature extraction model.

[0130] In this example, during practical application, the transient electromagnetic response signals at different well depths in a gas field were first measured, resulting in 3000 sets of field experimental data. These 3000 sets of data were then substituted into a pre-defined feature extraction model to obtain the calculated water intrusion level at the corresponding well depth. Analysis showed that the average prediction accuracy of these 3000 sample data was 86.45%, meeting a good standard for practical application. In other words, this feature extraction method can accurately and quantitatively calculate the water intrusion level at different well depths in different test wells, thereby effectively identifying formation-related information at different well depths and providing a good assessment of remaining oil and gas information, demonstrating promising application prospects.

[0131] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.

Claims

1. A method for feature extraction of downhole transient electromagnetic signals, characterized in that, include: Obtain the preset feature extraction model; The measured transient electromagnetic response signals at different depths in the well to be tested are obtained, and the target feature parameters at the corresponding depths are obtained based on the measured transient electromagnetic response signals. The target feature parameters are feature parameters whose correlation with the degree of water intrusion at different depths is greater than a preset correlation. The target feature parameters are substituted into the preset feature extraction model to obtain the terrain features at the corresponding well depth of the well to be measured; The method for obtaining the preset feature extraction model includes: Acquire standard transient electromagnetic response signals at different depths in a standard well; Calculate the characteristic parameters of the standard transient electromagnetic response signal in multiple dimensions; Obtain the correlation between each of the aforementioned feature parameters and the degree of water intrusion at different well depths, and select the feature parameters whose correlation with the degree of water intrusion is greater than the preset correlation as target feature parameters; The preset feature extraction model is established based on the target feature parameters; The acquisition of the correlation between each of the aforementioned feature parameters and the degree of water intrusion at different well depths includes: A preset correlation analysis algorithm is used to calculate the correlation between the characteristic parameters and the degree of water intrusion at different well depths.

2. The feature extraction method for downhole transient electromagnetic signals according to claim 1, characterized in that, Before calculating the feature parameters of the standard transient electromagnetic response signal in multiple dimensions, the method for obtaining the preset feature extraction model further includes: The standard transient electromagnetic response signal is preprocessed.

3. The feature extraction method for downhole transient electromagnetic signals according to claim 2, characterized in that, The data preprocessing of the standard transient electromagnetic response signal includes: data deduplication, missing value processing, outlier processing, and / or data noise reduction.

4. The feature extraction method for downhole transient electromagnetic signals according to claim 1, characterized in that, The multiple dimensions include statistical dimensions, time-frequency domain dimensions, shape metric dimensions, and / or hysteresis dimensions.

5. The feature extraction method for downhole transient electromagnetic signals according to claim 4, characterized in that, The feature values ​​of the statistical dimension include maximum value, minimum value, mean, variance, standard deviation, peak value, kurtosis and / or skewness; The eigenvalues ​​of the time-frequency domain dimension include the centroid frequency, average frequency, root mean square frequency, and / or frequency variance. The eigenvalues ​​of the shape metric dimension include Shannon entropy, Higitch fractal dimension, Katz fractal dimension, and / or generalized Hearst exponent; The eigenvalues ​​of the lag dimension include the autocorrelation eigenvalues ​​of the time series.

6. The feature extraction method for downhole transient electromagnetic signals according to claim 1, characterized in that, The step of establishing the preset feature extraction model based on the target feature parameters includes: The preset feature model is trained by using the target feature parameters as input and the water intrusion degree corresponding to different well depths as output. Weights are assigned to the target feature parameters to establish the preset feature extraction model.

7. The feature extraction method for downhole transient electromagnetic signals according to claim 6, characterized in that, The method of assigning weights to the target feature parameters includes: An adaptive enhancement ensemble algorithm is used to assign weights to the target feature parameters.

8. The feature extraction method for downhole transient electromagnetic signals according to claim 1, characterized in that, The method of establishing the preset feature extraction model based on the target feature parameters further includes: A regression random forest algorithm was used to obtain a nonlinear relationship data model between the standard transient electromagnetic response signal and the degree of water intrusion at different well depths by fitting a classification regression tree. The nonlinear relational data model is used as the preset feature extraction model.

Images