Automatic lifting inclinometry traction curve retardation characteristic positioning method

Abstract

The invention discloses an automatic lifting inclinometry traction curve retardation characteristic positioning method, and relates to the field of underground measurement, and the method comprises the steps: collecting original measurement data in a lifting inclinometry process, the original measurement data comprises an original traction force sequence, an original depth sequence and an environment parameter set, preprocessing the original traction force sequence to obtain a traction force fluctuation sequence; based on the environment parameter set, modulating a standard theoretical fingerprint of a preset single retardation type, and constructing a dynamic feature fingerprint database; analyzing the traction fluctuation sequence to detect and position an abnormal well section; constructing a linear hybrid model based on the observation signal extracted from the abnormal well section and the dynamic feature fingerprint database; carrying out constraint optimization inversion solution on the linear mixed model to obtain an optimal contribution degree weight coefficient; and outputting an analysis result containing the depth position information of the abnormal well section and the contribution degree information of each single retardation type in the abnormal well section. According to the method, automatic and refined identification and interpretation of underground complex retardation causes are realized.

Description

Technical Field

[0001] This invention relates to the field of downhole measurement, specifically to an automatic lifting and tilting traction curve retardation feature localization method. Background Technology

[0002] In oil and gas drilling, geological exploration, and engineering logging operations, pull-up logging is a crucial method for obtaining key wellbore trajectory parameters (inclination angle, azimuth angle) and assessing downhole conditions. During this process, by analyzing the traction curve (i.e., the traction curve) of the sensing cable as the downhole instrument is pulled up, the frictional condition and potential blockage points in the wellbore can be inferred. Accurately identifying and locating these blockage characteristics is of paramount engineering significance for assessing complex downhole conditions (such as keyways, cuttings bed accumulation, localized narrowing, excessive doglegs, etc.), optimizing drilling techniques, preventing stuck pipe accidents, and ensuring operational safety. Currently, the analysis of blockage characteristics in pull-up logging traction curves mainly relies on the experience of engineering technicians. Operators manually identify possible abnormal well sections and infer their causes by observing abrupt changes in curve shape, abnormal trends, or comparing them with historical curves. While this manual analysis method is direct, it has limitations such as strong subjectivity, low efficiency, and difficulty in real-time processing and analysis. Furthermore, when multiple hindering factors (such as dogleg friction and cuttings bed accumulation) act together in adjacent well sections, their signals superimpose and couple on the curve, forming complex composite anomaly waveforms. Existing methods can usually only make a general qualitative judgment on the entire composite anomaly (such as "high friction here"), but cannot quantitatively analyze the contribution ratio of each individual hindering type, making it difficult to accurately guide subsequent specific engineering decisions such as "should we prioritize wellbore drilling or deviation correction?"

[0003] In recent years, with the development of digital logging technology, some automated signal processing and pattern recognition methods have been introduced to assist or replace some manual analysis work. For example, alarms are triggered by tension mutations by setting thresholds, or statistical methods are used to identify baseline deviation signals. However, these methods mostly focus on judging the "presence of anomalies" and "rough location," and have not yet solved the deep-seated technical problem of separating and quantitatively analyzing the causes of complex resistance features. Therefore, developing an analytical method that can automatically and accurately locate resistance features and further intelligently analyze the contribution of multiple physical causes within them has become an urgent technical requirement for improving the accuracy and intelligence of downhole condition diagnosis. Summary of the Invention

[0004] Based on the shortcomings of the prior art described above, the purpose of this invention is to provide an automatic lifting and tilting traction curve stagnant feature positioning method to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an automatic lifting and tilting traction curve stagnant feature positioning method, comprising:

[0006] S1: Collect raw measurement data during the lifting and tilting process. The raw measurement data includes raw traction force sequence, raw depth sequence and environmental parameter set. The raw traction force sequence is preprocessed to obtain traction force fluctuation sequence.

[0007] S2: Based on the environmental parameter set, standard theoretical fingerprints of various preset single hindrance types are modulated to construct a dynamic feature fingerprint library that matches the current measurement conditions;

[0008] S3: Analyze the traction force fluctuation sequence to detect and locate abnormal well sections;

[0009] S4: Based on the observation signals and dynamic feature fingerprint database extracted from the abnormal well sections, a linear hybrid model is constructed to solve the contribution of each single hindrance type.

[0010] S5: Perform constrained optimization inversion on the linear hybrid model to obtain the optimal contribution weight coefficient of each single hindrance type in the abnormal well section;

[0011] S6: Outputs analysis results containing information on the depth and location of abnormal well sections, as well as the contribution information of each individual type of obstruction within them.

[0012] The present invention is further configured such that the set of environmental parameters includes drilling fluid density, plastic viscosity, yield value, bottom hole temperature, and bottom hole pressure.

[0013] The present invention is further configured such that S1 includes:

[0014] Outlier detection and removal are performed on the original traction force sequence and the original depth sequence, and the sequence after outlier removal is time-synchronized and aligned to obtain the processed traction force sequence and processed depth sequence on the spatiotemporal reference.

[0015] The processed traction force sequence is subjected to multi-scale discrete wavelet transform, and the high-frequency detail coefficients obtained by the transform are subjected to nonlinear soft thresholding using a preset threshold function. The denoised traction force sequence is then reconstructed by inverse wavelet transform.

[0016] Calculate the arithmetic mean of the denoised traction force sequence within the well section to be analyzed, and subtract the arithmetic mean from each data point of the denoised traction force sequence to obtain the traction force fluctuation sequence with zero mean.

[0017] The physical range validity of each parameter in the environmental parameter set is verified, and the parameters that pass the verification are normalized to form a standardized environmental parameter set.

[0018] The present invention is further configured such that S2 includes:

[0019] Obtain the standard theoretical fingerprint of each preset single hindrance type under standard reference conditions;

[0020] For each individual hysteresis type, the environmental coupling factor associated with the amplitude modulation of the hysteresis signal of that type is calculated based on the environmental parameter set;

[0021] Based on the sensor's sampling frequency and the cutoff frequency of the sensor's built-in hardware low-pass filter, the transfer function characterizing the instrument's measurement system response is determined, and this transfer function is transformed into the depth domain to obtain its corresponding discrete impulse response sequence.

[0022] The instrument response correction fingerprint is obtained by convolving the standard theoretical fingerprint of each single hindrance type with the discrete impulse response sequence.

[0023] Multiply the instrument response correction fingerprint with the corresponding environmental coupling factor to generate a dynamic feature fingerprint that matches the current measurement conditions;

[0024] Collect dynamic feature fingerprints corresponding to all single blockage types to construct a dynamic feature fingerprint library.

[0025] The present invention is further configured such that S3 includes:

[0026] Using a sliding window of preset length and step size, the traction force fluctuation sequence is traversed, and the local statistical characteristics of the data within each window are calculated. The local statistical characteristics include local standard deviation and local kurtosis.

[0027] Based on the local standard deviation sequence calculated from all windows, the median and absolute median of its statistical distribution are determined, and an adaptive anomaly detection threshold is set according to the median and absolute median.

[0028] The window center depth point with a local standard deviation greater than the adaptive anomaly detection threshold is marked as a candidate anomaly point, and the candidate anomaly points with continuous positions are aggregated to form a candidate anomaly segment.

[0029] Starting from the beginning and ending points of the candidate abnormal section, the search proceeds along the decreasing and increasing depth directions, respectively. The depth points where the local standard deviation first falls below the preset boundary judgment threshold are defined as the precise starting depth and precise ending depth of the abnormal well section, respectively.

[0030] Based on the depth range defined by the precise starting depth and precise ending depth, corresponding data points are extracted from the traction force fluctuation sequence to form an observation signal vector.

[0031] The present invention is further configured such that S4 includes:

[0032] Based on the depth range of the abnormal well section and the preset depth sampling interval, dynamic feature fingerprint segments of each single blockage type in the corresponding depth range are extracted from the dynamic feature fingerprint database.

[0033] Each extracted dynamic feature fingerprint segment is constructed into a fingerprint vector with the same length as the observed signal vector;

[0034] A linear mixture model is constructed, which represents the observed signal vector as the sum of a linear combination of fingerprint vectors and a residual error vector, where the linear coefficient of each fingerprint vector is the contribution weight coefficient of the single stagnation type corresponding to that fingerprint vector.

[0035] The linear hybrid model is represented as a matrix operation, where all fingerprint vectors are juxtaposed to form the design matrix, and all contribution weight coefficients to be determined are arranged to form the weight vector.

[0036] The present invention is further configured such that S5 includes:

[0037] A least-squares optimization problem is constructed with the objective function of minimizing the sum of squared residuals between the observed signal vector and the reconstructed signal calculated from the design matrix and weight vector.

[0038] Physical constraints are imposed on the least squares optimization problem, namely: all contribution weight coefficients in the weight vector must satisfy the non-negativity requirement, and their sum must satisfy the normalization requirement;

[0039] The least squares optimization problem with nonnegativity and normalization constraints is transformed into a standard form quadratic programming problem.

[0040] The interior point method is used to iteratively solve the quadratic programming problem to obtain the optimal weight vector that satisfies all constraints. Each element in the optimal weight vector is the optimal contribution weight coefficient corresponding to a single type of obstruction.

[0041] The present invention is further configured to calculate the reconstructed signal corresponding to the linear mixture model based on the optimal weight vector and the design matrix, and determine the goodness of fit by comparing the reconstructed signal with the observed signal vector, so as to verify the reliability of the linear mixture model and the inversion results.

[0042] The present invention is further configured such that S6 includes:

[0043] A depth location map is generated based on the traction force fluctuation sequence and the depth range of abnormal well sections;

[0044] Based on the optimal contribution weight coefficient, a contribution analysis diagram is generated for each single type of obstruction within each abnormal well section.

[0045] The present invention is further configured such that the preset single resistance type includes at least dogleg friction, local diameter reduction, rock cuttings bed accumulation, and keyway.

[0046] This invention provides an automatic pull-up inclinometer traction curve hindrance feature localization method. The method comprises: S1: collecting raw measurement data during the pull-up inclinometer process, including raw traction force sequence, raw depth sequence, and environmental parameter set; preprocessing the raw traction force sequence to obtain a traction force fluctuation sequence; S2: modulating standard theoretical fingerprints of various preset single hindrance types based on the environmental parameter set to construct a dynamic feature fingerprint database matching the current measurement conditions; S3: analyzing the traction force fluctuation sequence to detect and locate abnormal well sections; S4: constructing a linear hybrid model for solving the contribution of each single hindrance type based on the observation signals extracted from the abnormal well sections and the dynamic feature fingerprint database; S5: performing constrained optimization inversion on the linear hybrid model to obtain the optimal contribution weight coefficients of each single hindrance type in the abnormal well section; and S6: outputting the analysis results containing the depth and location information of the abnormal well section and the contribution information of each single hindrance type within it. The beneficial effects include:

[0047] 1. By calculating the environmental coupling factor and the instrument response transfer function respectively, the standard theoretical fingerprint is physically modulated to generate a dynamic feature fingerprint that strictly matches the current actual operating environment and instrument status. This process overcomes the systematic deviation caused by the mismatch between environmental parameters and instrument characteristics in the traditional fixed model, so that the feature matching benchmark can be dynamically adapted to the specific drilling fluid performance, downhole temperature and pressure and sensor dynamic characteristics, thereby improving the physical accuracy and model reliability of subsequent feature separation and quantitative analysis.

[0048] 2. By calculating local statistical features through a sliding window and dynamically setting an adaptive anomaly judgment threshold based on the statistical distribution of data across the entire well section, the traditional method of setting a fixed threshold based on human experience is replaced. This detection mechanism based on the characteristics of the data itself can automatically adapt to changes in background noise in different well sections, improve the detection sensitivity of weak and gradual anomaly signals, and further accurately locate the boundary by searching at both ends of the anomaly section, thus achieving objective and accurate determination of the start and end depths of the anomaly section and eliminating the subjectivity and inconsistency of manual interpretation.

[0049] 3. By characterizing the observed signal as a linear combination of multiple dynamic feature fingerprints and solving the contribution weight coefficients that satisfy the non-negativity and normalization constraints, the quantitative calculation of the contribution ratio of each single blockage type in the composite anomaly signal is realized. This technical solution breaks through the limitation of traditional methods that can only make qualitative judgments on presence or absence, and provides a technical means for decomposing the causes of downhole composite blockage and evaluating the quantitative contribution, so that the analysis results can leap from qualitative description to quantitative characterization.

[0050] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0052] Figure 1 The flowchart illustrates an automatic lifting and tilting traction curve stabilization feature positioning method as an exemplary embodiment of the present invention. Detailed Implementation

[0053] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0054] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0055] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0056] An automatic lifting and tilting traction curve stagnant feature localization method, such as Figure 1 As shown, it includes:

[0057] S1: Collect raw measurement data during the lifting and tilting process. The raw measurement data includes raw traction force sequence, raw depth sequence and environmental parameter set. The raw traction force sequence is preprocessed to obtain traction force fluctuation sequence.

[0058] S2: Based on the environmental parameter set, standard theoretical fingerprints of various preset single hindrance types are modulated to construct a dynamic feature fingerprint library that matches the current measurement conditions;

[0059] S3: Analyze the traction force fluctuation sequence to detect and locate abnormal well sections;

[0060] S4: Based on the observation signals and dynamic feature fingerprint database extracted from the abnormal well sections, a linear hybrid model is constructed to solve the contribution of each single hindrance type.

[0061] S5: Perform constrained optimization inversion on the linear hybrid model to obtain the optimal contribution weight coefficient of each single hindrance type in the abnormal well section;

[0062] S6: Outputs analysis results containing information on the depth and location of abnormal well sections, as well as the contribution information of each individual type of obstruction within them.

[0063] The present invention is further configured such that S1 includes:

[0064] Outlier detection and removal are performed on the original traction force sequence and the original depth sequence, and the sequence after outlier removal is time-synchronized and aligned to obtain the processed traction force sequence and processed depth sequence on the spatiotemporal reference.

[0065] The processed traction force sequence is subjected to multi-scale discrete wavelet transform, and the high-frequency detail coefficients obtained by the transform are subjected to nonlinear soft thresholding using a preset threshold function. The denoised traction force sequence is then reconstructed by inverse wavelet transform.

[0066] Calculate the arithmetic mean of the denoised traction force sequence within the well section to be analyzed, and subtract the arithmetic mean from each data point of the denoised traction force sequence to obtain the traction force fluctuation sequence with zero mean.

[0067] The physical range validity of each parameter in the environmental parameter set is verified, and the parameters that pass the verification are normalized to form a standardized environmental parameter set. The invention further specifies that the environmental parameter set includes drilling fluid density, plastic viscosity, yield value, bottom hole temperature, and bottom hole pressure. Specifically, a cable depth encoder collects and records instrument depth measurement data in real time to form an original depth sequence; a downhole tension sensor synchronously collects and records traction force measurement data during the lifting and tilting process to form a traction force sequence; a mud densitometer obtains drilling fluid density data; a rheometer obtains drilling fluid plastic viscosity and yield value data; a downhole temperature sensor or a geothermal gradient model obtains bottom hole temperature data; and a downhole pressure... Bottom-hole pressure data is acquired via pressure sensors or calculations based on hydraulic models. This data collectively constitutes the environmental parameter set. All data sources are synchronously acquired and recorded based on a unified time reference, ensuring that each data point from different sensors has a strictly corresponding timestamp, thereby guaranteeing the consistency of depth data, traction force data, and environmental parameter data in terms of spatiotemporal reference. A sliding window-based Hampel filter is used to detect and remove outliers from the original depth and traction force sequences. Outliers refer to abnormal data points caused by signal interruptions, pulse interference, etc. A preset fixed-length sliding window is defined centered on each data point in the original depth and traction force sequences, and the median and absolute median of all data within this window are calculated. If a data point's value deviates from the median of its window by more than a preset multiple of the absolute median difference, it is identified as an outlier. Linear interpolation is then performed using adjacent valid data points to fill the data gap at that location. After outlier removal, the processed depth sequence and traction force sequence are rigorously aligned and matched based on precise and unified timestamps to ensure that each depth measurement corresponds to a unique and time-synchronized traction force measurement. Simultaneously, the environmental parameter set data in time series format is matched with the aligned depth sequence to form an environmental parameter vector indexed by depth values. The cleaned and aligned traction force sequence is then decomposed into multiple scales using discrete wavelet transform, selecting preset wavelet basis functions such as Da... The ubechies4 wavelet is used to decompose the original traction force signal into approximation coefficients and detail coefficients on different frequency sub-bands. For the high-frequency detail coefficients obtained from the decomposition, an adaptive thresholding algorithm based on the Stein unbiased risk estimation principle is used for soft thresholding to reduce noise, thereby suppressing high-frequency random noise components in the signal. This process calculates a threshold independently for each level of detail coefficients, shrinking detail coefficients with absolute values ​​below the threshold to zero, and reducing the value of detail coefficients with absolute values ​​above the threshold to zero by the amount of the threshold. Finally, the thresholded detail coefficients of each level are combined with the approximation coefficients of the lowest frequency sub-band and reconstructed by inverse wavelet transform to obtain the denoised traction force sequence after removing high-frequency noise.The arithmetic mean of the denoised traction force sequence is calculated over the entire depth range of the target well section to be analyzed. This arithmetic mean is then subtracted from the value of each data point in the denoised traction force sequence to obtain a new sequence with a mean of zero, which is the traction force fluctuation sequence. This operation eliminates the influence of the overall average background force during the pulling operation, ensuring that the fluctuations in the sequence mainly reflect the relative tension changes caused by changes in wellbore contact conditions or geometric anomalies. The physical range validity of each parameter data in the depth-indexed environmental parameter vector is verified, and abnormal data points that significantly exceed the reasonable physical range, such as negative density values, are removed. Linear interpolation methods using neighboring valid data can be used to fill in the removed points. The verified environmental parameter data are then normalized, that is, each parameter's data is linearly mapped to a closed interval of zero and one, eliminating differences in the dimensions and orders of magnitude of different environmental parameters. This facilitates unified use and calculation by the subsequent physical model, forming a standardized environmental parameter set.

[0068] The present invention is further configured such that S2 includes:

[0069] The invention further specifies that the preset single resistance type includes at least dogleg friction, local narrowing, rock cuttings bed accumulation, and keyway.

[0070] For each individual hysteresis type, the environmental coupling factor associated with the amplitude modulation of the hysteresis signal of that type is calculated based on the environmental parameter set;

[0071] Based on the sensor's sampling frequency and the cutoff frequency of the sensor's built-in hardware low-pass filter, the transfer function characterizing the instrument's measurement system response is determined, and this transfer function is transformed into the depth domain to obtain its corresponding discrete impulse response sequence.

[0072] The instrument response correction fingerprint is obtained by convolving the standard theoretical fingerprint of each single hindrance type with the discrete impulse response sequence.

[0073] Multiply the instrument response correction fingerprint with the corresponding environmental coupling factor to generate a dynamic feature fingerprint that matches the current measurement conditions;

[0074] A dynamic feature fingerprint database is constructed by collecting dynamic feature fingerprints corresponding to all single resistance types. Specifically, for each preset single resistance type, its theoretical fingerprint is calculated under predefined standard reference conditions using numerical simulation or analytical modeling methods based on physical mechanisms. The standard reference conditions are a set of idealized physical and technical conditions set to establish a unified and reproducible analytical benchmark. Specifically, these include: setting the drilling fluid environment as a homogeneous and isothermal Newtonian fluid with preset constant standard values ​​for density, plastic viscosity, and yield value; maintaining the bottom hole temperature and pressure at standard ambient temperature and standard atmospheric pressure, i.e., 25 degrees Celsius and 1 standard atmosphere, respectively; and setting the instrument measurement... The system is simplified to an ideal linear system with infinite bandwidth, zero phase hysteresis, and unity gain. The purpose of this standard reference condition is to eliminate interference caused by fluid nonlinearity, temperature and pressure variations, and sensor dynamics in actual operations. Under this standard reference condition, corresponding theoretical models are established for each preset single resistance type, such as dogleg friction, local diameter reduction, cuttings bed accumulation, and keyways, based on their respective unique physical mechanisms. Specifically, the dogleg friction model is based on the contact mechanics of the tubing string and wellbore in a curved section; the local diameter reduction model is based on the fluid dynamics and solid contact mechanics of fluid flowing through a variable diameter channel; and the cuttings bed accumulation model is based on the mechanics of particulate matter and the principle of flow resistance. The keyway model is based on the geometric interference and tribological principles of the tubing string and irregular wellbore. Numerical simulation methods such as finite element analysis or finite difference calculations are used, or analytical mathematical formulas that have undergone rigorous theoretical derivation and verification are applied, to simulate and calculate the theoretical traction force variation curve along the well depth direction when various types of resistance are applied to the theoretical model with a specifically defined, physically meaningful benchmark unit strength. The benchmark unit strength is a standardized action quantity defined for quantitative characterization and comparison, such as the unit wellbore curvature change corresponding to dogleg friction, the unit flow cross-sectional area reduction rate corresponding to local diameter reduction, and the unit linear accumulation thickness corresponding to cuttings bed accumulation. And the unit geometric groove depth corresponding to the keyway; through the above calculations, the intrinsic response curves of each single hindrance type under idealized, interference-free conditions are obtained. This curve is defined as the standard theoretical fingerprint of the single hindrance type. The standard theoretical fingerprint characterizes the inherent, repeatable theoretical signal shape and amplitude characteristics of various hindrances after filtering out the influence of actual environmental fluctuations and the dynamic characteristics of the instrument measurement system. For each single hindrance type, based on the environmental parameter set, the environmental coupling factor associated with the amplitude modulation of the hindrance signal of that type is calculated. This environmental coupling factor is used to quantify the modulation effect of the current actual downhole environment relative to the standard reference conditions on the amplitude of a specific hindrance signal.The specific calculation process of the environmental coupling factor needs to comprehensively consider the influence of drilling fluid properties, downhole temperature, and bottom hole pressure. For each type of resistance, empirical or semi-empirical correction relationships are established for its signal amplitude as a function of drilling fluid density, plastic viscosity, and yield value. A correction relationship is also established for its signal amplitude as a function of bottom hole temperature, typically in the form of an exponential function. The key temperature sensitivity coefficient needs to be obtained through laboratory calibration. Simultaneously, a correction relationship is established for its signal amplitude as a function of bottom hole pressure. The correction coefficients calculated based on drilling fluid parameters, temperature parameters, and pressure parameters are then multiplied together to obtain a comprehensive reflection of the current environmental impact. The environmental coupling factor for this type of hindrance is used to characterize the proportion by which the actual downhole environment enhances or weakens this type of hindrance signal relative to its standard theoretical fingerprint. Then, the downhole tension sensor and its signal conditioning circuit are considered as a dynamic system, whose input-output characteristics are described by a transfer function. This dynamic system is simplified to a second-order low-pass filter model for characterization. The key parameters required to construct this transfer function are directly derived from the inherent characteristics of the sensor system itself, namely the signal sampling frequency and the cutoff frequency of the sensor's built-in hardware low-pass filter. Using the cutoff frequency, the natural frequency of the second-order low-pass filter is calculated according to the ratio of angular frequency to frequency. Simultaneously, based on... The typical dynamic response characteristics of the sensor system are considered, and its damping ratio parameter is preset to a specific value within a reasonable engineering range. The value range of the damping ratio parameter is 0.6 to 1.2. In this embodiment, the damping ratio parameter can be preset to 0.7, 1.0, or 1.2. Based on the calculated natural frequency and the preset damping ratio parameter, the transfer function of the second-order low-pass filter in the continuous time domain is constructed. The transfer function in the continuous time domain is converted into a digital filter transfer function in the discrete time domain using the bilinear transformation method. By performing an inverse transformation on the digital filter transfer function, its unit impulse response sequence is obtained using long division or recursive calculation. The preset depth for the inclined plane measurement operation is then determined. The sampling interval maps the unit impulse response sequence in the time domain to a time-depth correspondence, transforming it into a discrete impulse response sequence in the depth domain. This discrete impulse response sequence in the depth domain fully characterizes the shaping and filtering effects of the dynamic characteristics of the measurement system on the ideal input signal. A discrete convolution operation is then performed between the standard theoretical fingerprint of each single hysteresis type and the discrete impulse response sequence in the depth domain. This operation simulates the waveform change of the ideal standard theoretical fingerprint after passing through the dynamic characteristics of the actual measurement system, yielding the instrument response correction fingerprint. This operation applies a low-pass filter to the standard theoretical fingerprint that matches the actual frequency response characteristics of the instrument, making its waveform and frequency band closer to the actual observable signal of the sensor system.Each type of instrument response correction fingerprint is multiplied point-by-point with its corresponding depth-varying environmental coupling factor sequence. This multiplication operation superimposes the modulation effect of the environment on the signal amplitude onto the fingerprint waveform corrected by the instrument's dynamic characteristics, generating a dynamic feature fingerprint that matches the current measurement environment and instrument measurement system conditions. The dynamic feature fingerprints generated by four preset single-impedance types—dogleg friction, local diameter reduction, rock debris bed accumulation, and keyway—are aggregated to form a depth-indexed feature fingerprint set, i.e., the dynamic feature fingerprint library.

[0075] The present invention is further configured such that S3 includes:

[0076] Using a sliding window of preset length and step size, the traction force fluctuation sequence is traversed, and the local statistical characteristics of the data within each window are calculated. The local statistical characteristics include local standard deviation and local kurtosis.

[0077] Based on the local standard deviation sequence calculated from all windows, the median and absolute median of its statistical distribution are determined, and an adaptive anomaly detection threshold is set according to the median and absolute median.

[0078] The window center depth point with a local standard deviation greater than the adaptive anomaly detection threshold is marked as a candidate anomaly point, and the candidate anomaly points with continuous positions are aggregated to form a candidate anomaly segment.

[0079] Starting from the beginning and ending points of the candidate abnormal section, the search proceeds along the decreasing and increasing depth directions, respectively. The depth points where the local standard deviation first falls below the preset boundary judgment threshold are defined as the precise starting depth and precise ending depth of the abnormal well section, respectively.

[0080] Based on the depth interval defined by the precise starting and ending depths, corresponding data points are extracted from the traction force fluctuation sequence to form an observation signal vector. Specifically, a sliding window with a preset window length and sliding step size is first set. The window length must be greater than the expected width of a possible single simple anomaly signal, while being smaller than the length of a typical complex composite anomaly segment. The sliding step size is set to be smaller than the window length to ensure the continuity of analysis in the depth direction. Starting from the starting depth point of the traction force fluctuation sequence, the sliding window is gradually moved along the depth increasing direction with the sliding step size as the increment until the window covers the entire target well section to be analyzed. For each stop of the sliding window... For all traction force fluctuation data within a specific depth range covered by the location, local statistical characteristic calculations are performed to generate two key features. The first feature is the local standard deviation, obtained by calculating the root mean square of the sum of squared deviations of all data points within the window from the mean of the data within the window. This statistic is used to quantify the dispersion or strength of fluctuations of the data points within the local depth range. The second feature is the local kurtosis, obtained by calculating the ratio of the fourth central moment of the data within the window to the fourth power of its standard deviation, and subtracting the theoretical baseline value of the kurtosis of the standard normal distribution. This statistic is used to quantify whether the distribution shape of the local data is sharper or flatter than the standard normal distribution. A high local kurtosis value typically indicates the presence of sharp pulses or steep abrupt changes within that depth range. After traversing the entire well section, two new sequences strictly corresponding to the depth positions are obtained: the local standard deviation sequence and the local kurtosis sequence. Secondly, based on the local standard deviation sequence, a robust statistical method is used to estimate the background noise level. The median of this sequence is calculated to represent its central tendency, and its absolute median, i.e., the median of the absolute values ​​of the differences between all data points and the median, is calculated to measure the dispersion of the sequence. Based on the statistical relationship between the distribution range and the absolute median of approximately normal distribution data, an adaptive multiplier is calculated using the median and absolute median combined with empirically preset multipliers. The anomaly detection threshold is specifically set as the median value plus the product of the multiplier coefficient and the absolute median difference. This method can dynamically set the anomaly detection threshold based on the statistical characteristics of the data itself to adapt to the background noise changes in different well sections. Then, the local standard deviation sequence is traversed, and the data points whose local standard deviation values ​​exceed the adaptive anomaly detection threshold are marked as candidate anomaly points at the center depth of the window. These candidate anomaly points indicate potential anomaly locations where the traction fluctuation is significantly higher than the background noise level. In the depth direction, continuous or spaced candidate anomaly points with intervals smaller than the preset aggregation tolerance are merged to form continuous candidate anomaly segments. Each segment represents a potential anomaly well section that may contain stagnation characteristics.Then, for each candidate anomaly segment, precise boundary positioning is performed. The search proceeds from its starting endpoint in a decreasing depth direction, and simultaneously from its ending endpoint in a increasing depth direction. During the search, the local standard deviation value corresponding to each depth point is checked. When the local standard deviation value is first encountered to be lower than a preset boundary judgment threshold, the search stops and that depth point is recorded as the precise boundary. The boundary judgment threshold is typically set as a preset fixed proportion of the adaptive anomaly judgment threshold. Finally, based on the determined precise starting and ending depths of the anomaly segment, all data points within that depth interval are extracted from the traction force fluctuation sequence at preset depth sampling intervals. These data points are arranged in depth order to construct a column vector, which is the observation signal vector.

[0081] The present invention is further configured such that S4 includes:

[0082] Based on the depth range of the abnormal well section and the preset depth sampling interval, dynamic feature fingerprint segments of each single blockage type in the corresponding depth range are extracted from the dynamic feature fingerprint database.

[0083] Each extracted dynamic feature fingerprint segment is constructed into a fingerprint vector with the same length as the observed signal vector;

[0084] A linear mixture model is constructed, which represents the observed signal vector as the sum of a linear combination of fingerprint vectors and a residual error vector, where the linear coefficient of each fingerprint vector is the contribution weight coefficient of the single stagnation type corresponding to that fingerprint vector.

[0085] The linear hybrid model is represented as a matrix operation, where all fingerprint vectors are juxtaposed to form the design matrix, and all contribution weight coefficients are arranged to form the weight vector. Specifically, firstly, based on the precise start depth, precise end depth, and preset depth sampling interval of the abnormal well section, dynamic feature fingerprint data segments for each preset single hindrance type within the depth range are extracted from the dynamic feature fingerprint database. For each single hindrance type, based on the same depth start point, depth end point, and sampling interval as the abnormal well section, corresponding data segments are extracted from its continuous dynamic feature fingerprint curve. Each extracted data segment is constructed into a mathematical column vector in ascending order of depth; this column vector is the fingerprint corresponding to that hindrance type. The fingerprint vectors are aligned with the observation signal vectors extracted from the same anomalous well section, ensuring that the length of each fingerprint vector is exactly the same as the length of the observation signal vector. This guarantees the comparability of all vectors in the data dimension and satisfies the conditions for subsequent linear operations. This operation achieves strict spatial alignment of each fingerprint vector with the observation signal vector at the depth scale. Secondly, a linear mixture model is constructed based on the physical assumption that signals can be linearly superimposed. This linear mixture model represents the observation signal vector as the sum of a linear combination of all fingerprint vectors and a residual error vector. Its basic principle is that the complex anomalous signals actually observed are considered as the result of the superposition of standard signals generated by multiple known single blocking mechanisms at different intensities, combined with residual factors not covered by the model. In the model, the scalar coefficient multiplied with each fingerprint vector is the contribution weight coefficient to be determined. This contribution weight coefficient physically represents the contribution intensity or proportion of the specific hindrance type to the observed anomalous signal. The residual error vector is used to accommodate measurement noise, model error, and other unmodeled interference signals. It is usually assumed that the mathematical expectation of the residual error vector is zero. Then, the linear hybrid model is transformed into a matrix operation form to adapt to numerical optimization solution. Specifically, all the constructed fingerprint vectors are used as column vectors and arranged from left to right according to the preset hindrance type order to form a design matrix. Each column of the design matrix corresponds to a fingerprint pattern of a specific hindrance type, and each row corresponds to the same depth sampling point. Simultaneously, all the contribution weight coefficients to be determined are arranged from top to bottom in the same order as the fingerprint vector column to form a weight column vector. Through the above matrix construction, the observed signal vector is expressed as the product of the design matrix and the weight column vector, plus the residual error vector. This matrix expression transforms the engineering signal decomposition problem into a standard linear inverse problem. Finally, by performing the above fingerprint vector extraction, linear hybrid model construction, and matrix expression steps, a quantitative linear hybrid model connecting the known dynamic feature fingerprint database and the actual observed signal is established. This linear hybrid model transforms the problem of identifying downhole composite hindrance features into a signal decomposition and parameter inversion problem with the goal of solving the contribution weight coefficients of each hindrance type.By solving this linear hybrid model, the optimal weight vector is obtained, and the values ​​of each component directly quantify the contribution ratio of each preset single hindrance type in the current abnormal well section signal. This completes the quantitative mapping and analysis from the hybrid measurement signal to its internal physical composition ratio.

[0086] The present invention is further configured such that S5 includes:

[0087] A least-squares optimization problem is constructed with the objective function of minimizing the sum of squared residuals between the observed signal vector and the reconstructed signal calculated from the design matrix and weight vector.

[0088] Physical constraints are imposed on the least squares optimization problem, namely: all contribution weight coefficients in the weight vector must satisfy the non-negativity requirement, and their sum must satisfy the normalization requirement;

[0089] The least squares optimization problem with nonnegativity and normalization constraints is transformed into a standard form quadratic programming problem.

[0090] An interior-point method is used to iteratively solve the quadratic programming problem to obtain the optimal weight vector that satisfies all constraints. Each element in the optimal weight vector is the optimal contribution weight coefficient corresponding to a single hindrance type. The invention further involves calculating the reconstructed signal corresponding to the linear mixture model based on the optimal weight vector and the design matrix, and determining the goodness of fit by comparing the reconstructed signal with the observed signal vectors to verify the reliability of the linear mixture model and the inversion results. Specifically, an optimization problem based on the least squares principle is first constructed, aiming to solve for a set of optimal contribution weight coefficients such that the theoretical reconstructed signal generated by linearly combining the fingerprint vectors represented by the design matrix using these contribution weight coefficients is consistent with the signal obtained from measured data. The overall difference between the extracted observation signals is minimized. This overall difference is quantitatively characterized by calculating the square of the difference between the observed signal vector and the reconstructed signal vector at each sampling point at the same depth, and summing the squared differences over all depth points. The resulting sum is the residual sum of squares. Therefore, minimizing this residual sum of squares is set as the objective function to be minimized in this optimization problem. Secondly, two constraints derived from the physical background of the problem are imposed on this objective function: the first constraint requires all weight coefficients to be greater than or equal to zero, i.e., a non-negativity constraint. The physical basis of this constraint is that the contribution of any pre-defined single type of hindrance to the observed composite anomalous signal should, in a physical sense, be represented as a non-negative quantity. The contribution of the first constraint lacks a reasonable physical explanation; the second constraint requires that the sum of all weight coefficients must be strictly equal to one, i.e., a normalization constraint; this constraint explicitly defines the weight coefficients obtained by solving as the relative contribution ratio of each hindrance type, thereby ensuring that the sum of the total contributions of all identified hindrance types to the current composite anomaly signal is 100%, so that the final inversion result has a clear and interpretable physical meaning; then, the least squares optimization problem with non-negativity constraints and normalization constraints is equivalently transformed into the standard quadratic programming problem form in mathematical optimization theory; the characteristic of this quadratic programming problem is that its objective function is a quadratic function of the independent variable, and all its constraints are linear functions of the independent variable, wherein the independent variable is the contribution weight coefficient vector. For this type of quadratic programming problem, this embodiment uses the interior-point method. This algorithm transforms nonnegativity inequality constraints into equality constraints by introducing slack variables, thus reconstructing the original constrained optimization problem into an equality-constrained optimization problem. Based on this, an augmented Lagrangian function is constructed to transform the original constrained optimization problem into an unconstrained or equality-constrained extremum problem for solution. The specific numerical solution process adopts the Newton-Raphson iteration method. Each iteration requires constructing and solving the Karouch-Kun-Tucker linear equation system consisting of the objective function gradient, Hessian matrix, and constraint derivative matrix. By iteratively updating the solution vector, it gradually approaches the optimal point that satisfies the minimization of the objective function and all constraints. To control the solution process, a convergence tolerance and a maximum number of iterations are preset as termination conditions.When the iterative change of the solution vector is less than the convergence tolerance, the algorithm is deemed to have converged and the current solution is output as the optimal weight vector. If convergence is not achieved after the maximum number of iterations, the calculation is terminated and the current optimal solution is output. Finally, the algorithm outputs an optimal weight vector that satisfies all constraints, where each element corresponds to an optimal contribution weight coefficient for a pre-defined single hindrance type. Finally, the effectiveness of the optimal contribution weight coefficients obtained from the inversion solution and the linear mixture model is quantitatively verified. Based on the optimal weight vector and the design matrix, the reconstructed signal of the linear mixture model is calculated using matrix multiplication. The goodness of fit is calculated to objectively quantify the reconstructed signal and the observed signal. The goodness of fit is calculated as follows: The total variance of the observed signal is calculated, which is the sum of squares of the differences between all data points in the observed signal vector and their arithmetic mean; the residual variance of the linear mixture model is calculated, which is the sum of squares of the differences between the observed signal vector and the reconstructed signal vector at corresponding points; the goodness of fit is defined as the difference between the numerical value and the ratio of the residual variance to the total variance; the goodness of fit ranges from zero to one. The closer the value is to one, the higher the degree of fit between the reconstructed signal and the observed signal, that is, the more accurate the linear mixture model is in analyzing the causes of downhole composite hindrance characteristics, and the stronger the reliability of the inverted contribution weight coefficients.

[0091] The present invention is further configured such that S6 includes:

[0092] A depth location map is generated based on the traction force fluctuation sequence and the depth range of abnormal well sections;

[0093] Based on the optimal contribution weighting coefficient, contribution analysis maps corresponding to each single hindrance type within each abnormal well section are generated. Specifically, a depth location map is generated based on the traction force fluctuation sequence and the depth range of each abnormal well section. This depth location map plots the signal curve of the entire well section with depth as the vertical axis and traction force fluctuation value as the horizontal axis. The range and identification of each abnormal well section are highlighted on the curve with a bright background or marker lines, and the coordinate axis labels and scales are clearly marked. Based on the optimal contribution weighting coefficient, a contribution analysis map is generated for each abnormal well section. This contribution analysis map uses a pie chart or stacked bar chart to intuitively display the optimal contribution weighting coefficient of each single hindrance type in a proportional form, and the corresponding well section depth range is marked on the map. Finally, the depth location map and each contribution analysis map are systematically integrated to form a visual analysis report containing quantitative information on the spatial location and causal composition of composite hindrance characteristics.

[0094] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for automatically locating the resistance features of a traction curve during inclinometer lifting, characterized in that, include: S1: Collect raw measurement data during the lifting and tilting process. The raw measurement data includes raw traction force sequence, raw depth sequence and environmental parameter set. The raw traction force sequence is preprocessed to obtain traction force fluctuation sequence. S2: Based on the environmental parameter set, standard theoretical fingerprints of various preset single hindrance types are modulated to construct a dynamic feature fingerprint library that matches the current measurement conditions; S3: Analyze the traction force fluctuation sequence to detect and locate abnormal well sections; S4: Based on the observation signals and dynamic feature fingerprint database extracted from the abnormal well sections, a linear hybrid model is constructed to solve the contribution of each single hindrance type. S5: Perform constrained optimization inversion on the linear hybrid model to obtain the optimal contribution weight coefficient of each single hindrance type in the abnormal well section; S6: Outputs analysis results containing information on the depth and location of abnormal well sections, as well as the contribution information of each individual type of obstruction within them.

2. The automatic lifting and tilting traction curve resistance feature positioning method according to claim 1, characterized in that, The set of environmental parameters includes drilling fluid density, plastic viscosity, yield value, bottom hole temperature, and bottom hole pressure.

3. The automatic lifting and tilting traction curve resistance feature positioning method according to claim 1, characterized in that, S1 includes: Outlier detection and removal are performed on the original traction force sequence and the original depth sequence, and the sequence after outlier removal is time-synchronized and aligned to obtain the processed traction force sequence and processed depth sequence on the spatiotemporal reference. The processed traction force sequence is subjected to multi-scale discrete wavelet transform, and the high-frequency detail coefficients obtained by the transform are subjected to nonlinear soft thresholding using a preset threshold function. The denoised traction force sequence is then reconstructed by inverse wavelet transform. Calculate the arithmetic mean of the denoised traction force sequence within the well section to be analyzed, and subtract the arithmetic mean from each data point of the denoised traction force sequence to obtain the traction force fluctuation sequence with zero mean. The physical range validity of each parameter in the environmental parameter set is verified, and the parameters that pass the verification are normalized to form a standardized environmental parameter set.

4. The automatic lifting and tilting traction curve stagnant feature positioning method according to claim 1, characterized in that, S2 includes: Obtain the standard theoretical fingerprint of each preset single hindrance type under standard reference conditions; For each individual hysteresis type, the environmental coupling factor associated with the amplitude modulation of the hysteresis signal of that type is calculated based on the environmental parameter set; Based on the sensor's sampling frequency and the cutoff frequency of the sensor's built-in hardware low-pass filter, the transfer function characterizing the instrument's measurement system response is determined, and this transfer function is transformed into the depth domain to obtain its corresponding discrete impulse response sequence. The instrument response correction fingerprint is obtained by convolving the standard theoretical fingerprint of each single hindrance type with the discrete impulse response sequence. Multiply the instrument response correction fingerprint with the corresponding environmental coupling factor to generate a dynamic feature fingerprint that matches the current measurement conditions; Collect dynamic feature fingerprints corresponding to all single blockage types to construct a dynamic feature fingerprint library.

5. The automatic lifting and tilting traction curve resistance feature positioning method according to claim 1, characterized in that, S3 includes: Using a sliding window of preset length and step size, the traction force fluctuation sequence is traversed, and the local statistical characteristics of the data within each window are calculated. The local statistical characteristics include local standard deviation and local kurtosis. Based on the local standard deviation sequence calculated from all windows, the median and absolute median of its statistical distribution are determined, and an adaptive anomaly detection threshold is set according to the median and absolute median. The window center depth point with a local standard deviation greater than the adaptive anomaly detection threshold is marked as a candidate anomaly point, and the candidate anomaly points with continuous positions are aggregated to form a candidate anomaly segment. Starting from the beginning and ending points of the candidate abnormal section, the search proceeds along the decreasing and increasing depth directions, respectively. The depth points where the local standard deviation first falls below the preset boundary judgment threshold are defined as the precise starting depth and precise ending depth of the abnormal well section, respectively. Based on the depth range defined by the precise starting depth and precise ending depth, corresponding data points are extracted from the traction force fluctuation sequence to form an observation signal vector.

6. The automatic lifting and tilting traction curve stagnant feature positioning method according to claim 1, characterized in that, S4 includes: Based on the depth range of the abnormal well section and the preset depth sampling interval, dynamic feature fingerprint segments of each single blockage type in the corresponding depth range are extracted from the dynamic feature fingerprint database. Each extracted dynamic feature fingerprint segment is constructed into a fingerprint vector with the same length as the observed signal vector; A linear mixture model is constructed, which represents the observed signal vector as the sum of a linear combination of fingerprint vectors and a residual error vector, where the linear coefficient of each fingerprint vector is the contribution weight coefficient of the single stagnation type corresponding to that fingerprint vector. The linear hybrid model is represented as a matrix operation, where all fingerprint vectors are juxtaposed to form the design matrix, and all contribution weight coefficients to be determined are arranged to form the weight vector.

7. The automatic lifting and tilting traction curve resistance feature positioning method according to claim 1, characterized in that, S5 includes: A least-squares optimization problem is constructed with the objective function of minimizing the sum of squared residuals between the observed signal vector and the reconstructed signal calculated from the design matrix and weight vector. Physical constraints are imposed on the least squares optimization problem, namely: all contribution weight coefficients in the weight vector must satisfy the non-negativity requirement, and their sum must satisfy the normalization requirement; The least squares optimization problem with nonnegativity and normalization constraints is transformed into a standard form quadratic programming problem. The interior point method is used to iteratively solve the quadratic programming problem to obtain the optimal weight vector that satisfies all constraints. Each element in the optimal weight vector is the optimal contribution weight coefficient corresponding to a single type of obstruction.

8. The automatic lifting and tilting traction curve stagnant feature positioning method according to claim 7, characterized in that, Based on the optimal weight vector and design matrix, the reconstructed signal corresponding to the linear mixture model is calculated, and the goodness of fit is determined by comparing the reconstructed signal with the observed signal vector, so as to verify the reliability of the linear mixture model and the inversion results.

9. The automatic lifting and tilting traction curve resistance feature positioning method according to claim 1, characterized in that, S6 includes: A depth location map is generated based on the traction force fluctuation sequence and the depth range of abnormal well sections; Based on the optimal contribution weight coefficient, a contribution analysis diagram is generated for each single type of obstruction within each abnormal well section.

10. The automatic lifting and tilting traction curve stagnant feature positioning method according to claim 1, characterized in that, The preset single resistance type includes at least dogleg friction, local narrowing, rock debris bed accumulation, and keyway.

Images