Self-adaptive dynamic stratum identification method based on multi-source drilling parameters

Abstract

A self-adaptive dynamic stratum recognition method based on multi-source drilling parameters comprises the steps that firstly, the drilling rate, the rotating speed and oil pressure parameters are collected through a sensor set, a drilling depth-time curve is constructed, time sequence data are segmented through a sliding window, and filtering, standardization and abnormal value processing are carried out; calculating a mean value, a standard deviation and skewness statistical characteristic of each parameter in the window, and constructing a nine-dimensional characteristic vector in combination with an energy ratio and a covariant coefficient; updating a stratum feature library based on an incremental learning algorithm, and dynamically optimizing a clustering center to realize autonomous evolution; a confidence coefficient threshold value is set according to the statistical characteristics and the clustering distance, and stratum type dual discrimination is completed; by comparing a real-time judgment result with manual sampling data, the generalization of a learning rate parameter optimization model is dynamically adjusted; and outputting a stratum type corresponding to the current drilling depth and a confidence degree evaluation result based on the optimized clustering model. According to the invention, by improving a data processing mechanism and an adaptive decision rule, high-precision and strong-generalization real-time stratum identification is realized.

Description

Technical Field

[0001] The present invention belongs to the technical fields of geological exploration and geotechnical engineering, and particularly relates to an adaptive dynamic formation identification method based on multi-source drilling parameters. Background Art

[0002] In the field of geological exploration, the accurate identification of formation information is decisive for geological steering, engineering site selection, foundation evaluation, etc.

[0003] Traditional formation identification relies on manual core sampling and laboratory analysis, with low efficiency and no real-time feedback. Although automated technologies based on drilling parameters have developed, their core bottlenecks are still prominent, as follows:

[0004] First, the static threshold has poor adaptability. In the prior art, fixed thresholds are used to divide formation types, without fully considering the dynamic coupling effects of multiple parameters such as oil pressure and rotational speed. Such methods are difficult to adapt to the real-time changes of different hardness formations, resulting in a significant increase in the misjudgment rate of hard rocks and loose formations, and affecting the exploration accuracy.

[0005] Second, the formation classification system is single. Current methods are mostly limited to a single lithology classification standard, lacking multi-level quantitative descriptions of the formation weathering degree and having insufficient compatibility. This defect leads to limited resolution in the identification of complex interbedded formations and is difficult to meet the requirements of refined exploration.

[0006] Third, the model learning ability is fixed. Traditional algorithms adopt a fixed parameter update mechanism without introducing real-time feedback and dynamic learning ability. The model has poor adaptability in new geological environments or thin formations and requires frequent manual intervention for calibration, seriously restricting the engineering efficiency and automation level.

[0007] The above problems severely restrict the engineering universality of formation identification. Therefore, it is urgent to construct an adaptive identification system that integrates dynamic threshold adjustment, multi-standard compatible classification, and feedback-based incremental learning to break through the triple technical bottlenecks of real-time performance, generalization, and accuracy. Summary of the Invention

[0008] In order to overcome the deficiencies of the prior art, the present invention provides an adaptive dynamic formation identification method based on multi-source drilling parameters, which realizes high-precision and strong-generalization real-time formation identification by improving the data processing mechanism and adaptive decision rules.

[0009] The technical solution adopted by the present invention to solve its technical problems is:

[0010] An adaptive dynamic formation identification method based on multi-source drilling parameters, comprising the following steps:

[0011] Step 1, collect the mechanical parameters of the drilling rig in real time during the drilling process. The sensor group is at F SCollect the drilling rate R, rotational speed S, and drilling oil pressure P, and construct a time-series data stream {R t , S t , P t}, where R t is the drilling rate at the t-th moment, S t is the rotational speed at the t-th moment, P t is the drilling oil pressure at the t-th moment, t is a positive integer, and its value range is 1 to N. Calculate the borehole depth through displacement accumulation, and construct a borehole depth-time curve in combination with the time series;

[0012] Step 2: Perform sliding window segmentation. Define the dynamic window length L, where T is the average formation thickness. The window contains N data points, and set the sampling interval and step size;

[0013] Step 3: Perform data preprocessing on the acquisition parameters in Step 1. The process is as follows: Perform Savitzky-Golay filtering on the acquisition data points within the window length and perform Z-score normalization on the data within the window; then perform outlier removal. If a data point deviates from the mean by more than 3σ, mark it as an outlier and replace it with linear interpolation of the previous point. Finally, obtain the i-th data point X i after data processing. The formula is

[0014]

[0015] When N is odd, When N is even,

[0016] where, is the i-th data within the window after filtering, X t+n represents the data point n units after the i-th data within the window, n is a process coefficient, c n is the weight coefficient generated by local polynomial least squares fitting; μ is the mean of the data within the window; σ is the standard deviation of the data within the window.

[0017] Step 4: Start generating statistical eigenvalues within the window, and calculate the mean (μ R , μ S , μ P ), standard deviation (σ R , σ S , σ P ), and skewness (γ R , γ S , γ P ) of each parameter within the window. Among them, μ R is the mean of the drilling rate R, μ S is the mean of the rotational speed S, μ P is the mean of the drilling oil pressure P, σ Ris the standard deviation of the drilling rate R, σ S is the standard deviation of the rotational speed S, σ P is the standard deviation of the drilling oil pressure P, γ R is the skewness of the drilling rate R, γ S is the skewness of the rotational speed S, γ P is the skewness of the drilling oil pressure P;

[0018] Step 5 Calculate the engineering feature structure including the energy ratio and the covariance coefficient, the formulas are as follows

[0019]

[0020] where, E ratio is the energy ratio, R i is the drilling rate at the i-th moment, S i is the rotational speed at the i-th moment, P i is the drilling oil pressure at the i-th moment, C RS covariance coefficient;

[0021] Step 6 Construct the feature vector, combine the statistical and engineering features to form a 9-dimensional feature vector F = [μ R , σ R , γ R , μ S , σ S , γ S , E ratio , C RS , μ P ;

[0022] Step 7 Perform incremental clustering learning for dynamic adjustment, start initialization, set n initial clustering centers C0 according to the formation classification standard, and update online: after drilling L meters each time, calculate the Euclidean distance between the current feature vector F and each clustering center, and update the nearest neighbor center. The formula is as follows:

[0023]

[0024] where, α is the learning rate, dynamically adjust the formation feature library, C i new is the updated clustering center, C i old is the clustering center before update;

[0025] Step 8 Statistic the threshold, dynamically calculate the mean μ and the standard deviation σ based on the data within the window, and set the threshold interval [μ - kσ, μ + kσ], where k is adaptively adjusted according to the formation hardness;

[0026] Step 9 Model confidence threshold: According to the clustering distance d = ||F - C i ||, C iis the clustering center, and set the confidence level Conf = e -d / 2 , and when Conf < 0.7, trigger reclassification;

[0027] In step 10, start the primary determination. If the real-time parameters R, S, and P are all within the statistical thresholds and Conf < 0.7, directly output the formation type corresponding to the current clustering.

[0028] Furthermore, the method further includes the following steps:

[0029] In step 11, start the secondary determination. When the parameters exceed the statistical thresholds but Conf ≥ 0.5, start the multi-parameter fusion analysis:

[0030] Hard rock formation determination condition: R < μ R -1.5σ R and P > μ p +1.0σ p ;

[0031] Loose layer determination condition: R > μ R +2.0σ R and the S fluctuation coefficient γ S > 1.2;

[0032] Finally, perform feedback optimization. After every m meters of drilling, compare the determination result with the artificial sampling data, calculate the error rate η. When η ≤ 0.25, dynamically adjust the learning rate α: α = 0.1·(1 - 0.8η).

[0033] Still further, the process of step 1 is as follows:

[0034] In step 1.1, collect the drilling rate R, rotation speed S, and oil pressure P at a frequency of 10 Hz through the sensor group, and construct a time series data stream {R t , S t , P t};

[0035] In step 1.2, calculate the borehole depth through displacement accumulation, and construct a borehole depth-time curve, a rotation speed-depth curve, and an oil pressure-depth curve in combination with the time series.

[0036] Preferably, in step 2, define the dynamic window length L = max(5m, 2T), where T is the average formation thickness, the initial value of T = 1m, the window contains N = [L / 0.1] data points, 0.1m is the sampling interval, and the step size is 0.5T.

[0037] Even further, the process of step 3 is as follows:

[0038] In step 3.1, perform data preprocessing, and perform Savitzky-Golay filtering on R, S, and P respectively;

[0039] Step 3.2 Standardization: Perform Z-score normalization on the data within the window using the following formula

[0040]

[0041] where is the data within the window after filtering, i is the i-th data within the window; μ is the mean of the data within the window; σ is the standard deviation of the data within the window;

[0042] Step 3.3 Outlier removal: If a data point deviates from the mean by more than 3σ, mark it as an outlier and replace it with linear interpolation of the previous point.

[0043] In step 8 described above, k is adaptively adjusted according to the formation hardness, and the formula is as follows:

[0044]

[0045] where the higher the drilling oil pressure P, the harder the formation, and the smaller k is.

[0046] The beneficial effects of the present invention are mainly manifested in: achieving high-precision and strong generalization real-time formation identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a graph showing the variation of drilling parameters with time.

[0048] Figure 2 is a graph of drilling depth - time.

[0049] Figure 3 is a graph of rotational speed - depth.

[0050] Figure 4 is a graph of oil pressure - depth. DETAILED DESCRIPTION OF THE INVENTION

[0051] The present invention will be further described below with reference to the accompanying drawings.

[0052] Refer to Figures 1 to 3 , an adaptive dynamic formation identification method based on multi-source drilling parameters, comprising the following steps:

[0053] Step 1: Real-time collect the mechanical parameters of the drilling rig during the drilling process. The sensor group collects the drilling rate R, rotational speed S, and drilling oil pressure P at a frequency of F S to construct a time-series data stream {R t , S t , P t}, where R t is the drilling rate at the t-th moment, S t is the rotational speed at the t-th moment, and P tis the drilling oil pressure at the t-th moment, where t is a positive integer with a value range of 1 to N. The drilling depth is calculated by displacement accumulation, and a drilling depth-time curve is constructed by combining time series. The process is as follows:

[0054] Step 1.1: Collect the drilling rate R (m / min), rotational speed S (rpm), and oil pressure P (MPa) at a frequency of 10 Hz through a sensor group to construct a time-series data stream {R t , S t , P t};

[0055] Step 1.2: Calculate the drilling depth by displacement accumulation, and construct a drilling depth-time curve, a rotational speed-depth curve, and an oil pressure-depth curve by combining time series;

[0056] Step 2: Perform sliding window segmentation. Define the dynamic window length L = max(5m, 2T), where T is the average thickness of the formation, with an initial value of T = 1m. The window contains N = [L / 0.1] data points, 0.1m is the sampling interval, and the step size is (0.5T);

[0057] Step 3: Perform data preprocessing on the collected parameters in Step 1: Perform Savitzky-Golay filtering on the collected data points within the window length and perform Z-score normalization on the data within the window; then perform outlier removal. If a data point deviates from the mean by more than 3σ, it is marked as an outlier and replaced with linear interpolation of the previous point. Finally, obtain the i-th data point X i The formula is

[0058]

[0059] When N is odd, When N is even,

[0060] where is the i-th data within the filtered window, X t+n represents the data point n units after the i-th data within the window, n is a process coefficient, c n is the weight coefficient generated by local polynomial least squares fitting; μ is the mean of the data within the window; σ is the standard deviation of the data within the window;

[0061] The process of Step 3 is as follows:

[0062] Step 3.1: Perform data preprocessing, perform Savitzky-Golay filtering on R, S, and P respectively. In this embodiment, the window length = 11 and the polynomial order = 2 to eliminate high-frequency noise;

[0063] Step 3.2 Standardization: Perform Z-score normalization on the data within the window as follows

[0064]

[0065] where is the i-th data within the window after filtering; μ is the mean of the data within the window; σ is the standard deviation of the data within the window; X i is the i-th data point after data processing;

[0066] Step 3.3 Outlier removal: If the data point deviates from the mean by more than 3σ, mark it as an outlier and replace it with linear interpolation of the previous point;

[0067] Step 4 Start generating statistical feature values within the window, calculate the mean (μ R , μ S , μ P ), standard deviation (σ R , σ S , σ P ), and skewness (γ R , γ S , γ P ) of each parameter within the window. Among them, μ R is the mean of the drilling rate R, μ S is the mean of the rotational speed S, μ P is the mean of the drilling hydraulic pressure P, σ R is the standard deviation of the drilling rate R, σ S is the standard deviation of the rotational speed S, σ P is the standard deviation of the drilling hydraulic pressure P, γ R is the skewness of the drilling rate R, γ S is the skewness of the rotational speed S, γ P is the skewness of the drilling hydraulic pressure P;

[0068] Step 5 Calculate engineering feature construction including energy ratio and covariance coefficient, the formulas are as follows

[0069]

[0070] where E ratio is the energy ratio, R i is the drilling rate at the i-th moment, S i is the rotational speed at the i-th moment, P i is the drilling hydraulic pressure at the i-th moment, C RS is the covariance coefficient;

[0071] Step 6 Construct a feature vector, combine statistical and engineering features to form a 9-dimensional feature vector F = [μ R , σ R , γ R , μS , σ S , γ S , E ratio , C RS , μ P ;

[0072] Step 7 performs incremental clustering learning for dynamic adjustment. Start with initialization. Set 6 initial cluster centers C0 according to the formation classification standard (such as GB50021 - 2018) for online update: After drilling every L meters, calculate the Euclidean distance between the current feature vector F and each cluster center, and update the nearest neighbor center. The formula is as follows

[0073]

[0074] where α is the learning rate, which dynamically adjusts the formation feature library, and α = 0.1; C i new is the updated cluster center, and C i old is the cluster center before update. ;

[0075] Step 8 calculates the threshold. Dynamically calculate the mean μ and standard deviation σ based on the data within the window, and set the threshold interval [μ - kσ, μ + kσ], where k is adaptively adjusted according to the formation hardness. The formula is as follows:

[0076]

[0077] where the higher the drilling oil pressure, the harder the formation, and the smaller k is;

[0078] Step 9 model confidence threshold: According to the clustering distance d = ||F - C i ||, where C i is the cluster center, set the confidence Conf = e -d / 2 , and when Conf < 0.7, trigger re - classification;

[0079] Step 10 starts the primary determination. If the real - time parameters R, S, and P are all within the statistical threshold and Conf < 0.7, directly output the formation type corresponding to the current cluster;

[0080] Step 11 starts the secondary determination. When the parameters exceed the statistical threshold but Conf ≥ 0.5, start the multi - parameter fusion analysis:

[0081] Hard rock formation determination condition: R < μ R - 1.5σ R and P > μ p + 1.0σ p ;

[0082] Loose layer determination condition: R > μ R + 2.0σR and the S fluctuation coefficient γ S > 1.2;

[0083] Finally, perform feedback optimization. After every 10 meters of drilling, compare the determination result with the artificial sampling data, calculate the error rate η, and dynamically adjust the learning rate α: when η ≤ 0.25, α = 0.1·(1 - 0.8η).

[0084] Example 1, simulating the calculation of the dynamic threshold. Taking the sandy soil layer as an example, the process is as follows:

[0085] Step 1: Real-time collect the drilling rig mechanical parameters during the drilling process and set the window parameters. The process is as follows:

[0086] Step 1.1: Generate a data window. When drilling in the sandy soil layer, set L = 5m and calculate that there are N = 50 data points in the window.

[0087] Step 1.2: Through the real-time drilling process, obtain the sampling parameter range: R ∈ [1.2, 2.5] m / min, S ∈ [80, 120] rpm, P ∈ [8, 12] MPa.

[0088] Step 2: Perform feature calculation, μ R = 1.8, σ R = 0.3, E ratio = 18;

[0089] Step 3: Update the clustering center. The corresponding C for the sandy soil layer 砂土层 = [1.7, 0.35, 0.1, 95, 15, 0.3, 20, 5.2, 10];

[0090] Step 4: Statistically calculate the threshold The interval is [1.8 - 1.5×0.3, 1.8 + 1.5×0.3] = [1.35, 2.25] m / min;

[0091] Step 5: Calculate the confidence level, Conf = e -0.8 / 2 = 0.67;

[0092] Step 6: When the real-time R = 2.3 m / min (exceeding the threshold), but Conf = 0.67 ≥ 0.5, trigger a secondary determination;

[0093] Step 7: Since R > 2.25 and the S fluctuation coefficient γ s = 1.5 > 1.2, it is determined as a sandy gravel layer.

[0094] Example 2, simulating the adaptive weight adjustment. Taking the hard rock formation as an example, the process is as follows:

[0095] Step 1): Calculate the initial parameters. For example, in the hard rock formation, μ P= 18 MPa,

[0096] Step 2) If the misjudgment rate reaches η = 0.15 in the manual sampling verification, then update the learning rate to α = 0.1×(1 - 0.8×0.15) = 0.088.

[0097] The content described in the embodiments of this specification is only an enumeration of the implementation forms of the inventive concept and is only for illustrative purposes. The protection scope of the present invention should not be regarded as limited to the specific forms stated in this embodiment, and the protection scope of the present invention also extends to equivalent technical means that can be conceived by those of ordinary skill in the art based on the inventive concept of the present invention.

Claims

1. An adaptive dynamic formation identification method based on multi-source drilling parameters, characterized in that, The method includes the following steps: Step 1: Real-time acquisition of drilling rig mechanical parameters during drilling. The sensor group is based on F S The drilling rate R, speed S, and drilling oil pressure P are collected frequently to construct a time series data stream {R t ,S t ,P t }, where R t is the drilling rate at the tth moment, S t is the speed at the tth moment, P t is the drilling oil pressure at the tth moment, t is a positive integer, ranging from 1 to N. The drilling depth is calculated by accumulating displacement, and the drilling depth-time curve is constructed by combining the time series; In step 2, perform sliding window segmentation. Define the dynamic window length L, where T is the average formation thickness, the window contains N data points, and set the sampling interval and step size; Step 3 preprocesses the collected parameters in Step 1, and the process is as follows: perform Savitzky-Golay filtering on the collected data points within the window length and perform Z-score normalization on the data within the window; then perform outlier removal. If the data point deviates from the mean by more than 3σ, it is marked as an outlier and replaced with linear interpolation of the previous point. Finally, the i-th data point X after data processing is obtained i , and the formula is When N is odd, When N is even, Among them, is the i-th data within the filtered window, X t+n represents the data point n units after the i-th data within the window, n is the process coefficient, c n is the weight coefficient generated by local polynomial least squares fitting; μ is the mean of the data within the window; σ is the standard deviation of the data within the window; Step 4 starts to generate the statistical eigenvalues within the window and calculates the mean values (μ R , μ S , μ P ), standard deviations (σ R , σ S , σ P ), and skewness (γ R , γ S , γ P ) of the parameters within the window. Among them, μ R is the mean value of the drilling rate R, μ S is the mean value of the rotational speed S, μ P is the mean value of the drilling oil pressure P, σ R is the standard deviation of the drilling rate R, σ S is the standard deviation of the rotational speed S, σ P is the standard deviation of the drilling oil pressure P, γ R is the skewness of the drilling rate R, γ S is the skewness of the rotational speed S, γ P is the skewness of the drilling oil pressure P; In step 5, calculate the engineering feature construction including the energy ratio and covariance coefficient. The formulas are as follows Among them, E ratio is the energy ratio, R i is the drilling rate at the i-th moment, S i is the rotational speed at the i-th moment, P i is the drilling oil pressure at the i-th moment, C RS covariance coefficient; Step 6 Construct the feature vector, combine the statistical and engineering features to form a 9-dimensional feature vector F = [μ R , σ R , γ R , μ S , σ S , γ S , E ratio , C RS , μ P ; In step 7, perform incremental clustering learning for dynamic adjustment. Start initialization. Set n initial cluster centers C0 according to the formation classification standard for online update: After drilling L meters each time, calculate the Euclidean distance between the current feature vector F and each cluster center, and update the nearest neighbor center. The formula is as follows: Among them, α is the learning rate, which dynamically adjusts the formation feature library, and C i new is the updated cluster center, and C i old is the cluster center before update; In step 8, calculate the threshold. Dynamically calculate the mean μ and standard deviation σ based on the data within the window, and set the threshold interval [μ - kσ, μ + kσ], where k is adaptively adjusted according to the formation hardness; Step 9 Model confidence threshold: According to the clustering distance d = ||F - C i ||, where C i is the clustering center, set the confidence level, and trigger reclassification when Conf < 0.7; Step 10 starts the primary determination. If the real-time parameters R, S, and P are all within the statistical thresholds and Conf < 0.7, directly output the current clustering pair with Conf = e -d / 2 corresponding formation type.

2. The adaptive dynamic formation identification method based on multi-source drilling parameters according to claim 1, characterized in that The method further includes the following steps: In step 11, start secondary determination. When the parameter exceeds the statistical threshold but Conf≥0.5, initiate multi-parameter fusion analysis: Determination conditions for hard rock strata: R < μ R -1.5σ R and P > μ p +1.0σ p ; Loose layer determination condition: R > μ R +2.0σ R and the S fluctuation coefficient γ S > 1.2; Finally, perform feedback optimization. After drilling m meters each time, compare the determination result with the artificial sampling data, calculate the error rate η. When η≤0.25, dynamically adjust the learning rate α: α = 0.1·(1 - 0.8η).

3. The adaptive dynamic formation identification method based on multi-source drilling parameters according to claim 1 or 2, characterized in that, The process of step 1 is as follows: Step 1.1 Collect the drilling rate R, rotational speed S, and oil pressure P at a frequency of 10 Hz through the sensor group to construct a time-series data stream {R t , S t , P t}; In step 1.2, calculate the borehole depth through displacement accumulation, and construct the borehole depth-time curve, rotation speed-depth curve, and oil pressure-depth curve in combination with the time series.

4. An adaptive dynamic formation identification method based on multi-source drilling parameters according to claim 1 or 2, characterized in that, In step 2, define the dynamic window length L = max(5m, 2T), where T is the average formation thickness, the initial value of T is 1m, the window contains N = [L / 0.1] data points, 0.1m is the sampling interval, and the step size is 0.5T.

5. An adaptive dynamic formation identification method based on multi-source drilling parameters according to claim 1 or 2, characterized in that The process of step 3 is as follows: In step 3.1, perform data preprocessing, and perform Savitzky-Golay filtering on R, S, and P respectively; In step 3.2, standardize: Perform Z-score normalization on the data within the window. The formula is as follows Among them, is the i-th data within the filtered window; μ is the mean of the data within the window; σ is the standard deviation of the data within the window; X i is the i-th data point after data processing; In step 3.3, remove outliers. If a data point deviates from the mean by more than 3σ, mark it as an outlier and replace it with linear interpolation of the previous point.

6. The adaptive dynamic formation identification method based on multi-source drilling parameters according to claim 1 or 2, characterized in that, In step 8, k is adaptively adjusted according to the formation hardness. The formula is as follows: Among them, the higher the drilling oil pressure P, the harder the formation, and the smaller k is.

Images