A downhole pulse generation control system based on pressure feedback compensation
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XINGTAI QIAODONG NONGZIGONGSIPING TOWNSHIP COUNTY AGRI SUPPLIES SALES OU TLET
- Filing Date
- 2026-05-19
- Publication Date
- 2026-08-07
AI Technical Summary
[0004]为了弥补以上不足，本发明提供了一种基于压力反馈补偿的井下脉冲发生控制系统，旨在改善传统的脉冲触发方式大都采用固定阈值或仅依赖原始压力信号幅值比较，容易造成频繁误触发或漏触发的问题
[0053]1. In this invention, the prediction triggering module uses a prediction model to generate predicted values for low-dimensional feature data and calculates feature residuals. Then, it combines the feature residuals with a dynamic threshold constructed based on the statistical features of the pressure signal to determine whether to trigger pulse transmission. This improves the problem that traditional pulse triggering methods mostly use fixed thresholds or rely solely on the comparison of the amplitude of the original pressure signal. Due to the strong downhole pressure noise and drastic changes in operating conditions, this often results in frequent false triggering or missed triggering.
Smart Images
Figure CN122533558A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of downhole measurement while drilling communication technology, and in particular to a downhole pulse generation and control system based on pressure feedback compensation. Background Technology
[0002] In measurement-while-drilling (MSD) and logging-while-drilling (LOD) systems, engineering parameters and geological data acquired by downhole tools are typically transmitted to the surface in the form of pressure pulses via mud pulse generators. The pulse generation control method directly affects the real-time performance and reliability of downhole data. In existing technologies, downhole pulse triggering often employs fixed pressure thresholds or methods based on comparisons of the original pressure signal amplitude. Some systems have introduced simple adaptive thresholds or filtering processes.
[0003] Traditional pulse triggering methods mostly use fixed thresholds or rely solely on the comparison of the amplitude of the original pressure signal. Due to the strong downhole pressure noise and drastic changes in operating conditions, this often leads to frequent false triggering or missed triggering. Summary of the Invention
[0004] To overcome the above shortcomings, this invention provides a downhole pulse generation control system based on pressure feedback compensation, which aims to improve the problem that traditional pulse triggering methods mostly use fixed thresholds or rely solely on the comparison of the amplitude of the original pressure signal, which easily leads to frequent false triggering or missed triggering.
[0005] This invention provides the following technical solution: a downhole pulse generation and control system based on pressure feedback compensation includes:
[0006] The multi-source preprocessing module is used to collect downhole multi-source time-series data, preprocess the multi-source time-series data, and construct a unified state vector based on the preprocessed multi-source time-series data.
[0007] The feature dimensionality reduction module is used to perform low-complexity feature extraction and data dimensionality reduction on a unified state vector. The feature extraction methods include sliding window statistics, temporal difference, and low-order wavelet packet decomposition to generate low-dimensional feature data.
[0008] The prediction triggering module is used to acquire downhole pressure signals in real time, extract statistical features of the pressure signals, construct dynamic thresholds based on the statistical features of the pressure signals, generate predicted values for low-dimensional feature data using a prediction model, calculate feature residuals, and combine feature residuals with dynamic thresholds to determine whether to trigger pulse transmission.
[0009] The graded compensation module is used to perform graded judgment based on the feature residual of low-dimensional feature data after the pulse transmission trigger takes effect. Based on the graded judgment result, it selects to output a standard pulse or perform pressure feedback compensation modulation on the pulse. When the noise intensity of the pressure signal exceeds the noise intensity limit value determined based on the historical quantile, or the pressure variance exceeds the pressure variance limit value determined based on the historical quantile, the reference pulse is fixedly output.
[0010] The closed-loop iteration module is used to collect real-time downhole signals after pulse output to construct evaluation indicators. Based on low-dimensional feature data and evaluation indicators, iteratively update the prediction model parameters, dynamic threshold parameters, and pulse compensation parameters. Boundary constraints are set during the iterative update process to complete the closed-loop control of the entire process.
[0011] By adopting the above technical solution, the prediction triggering module uses the prediction model to generate predicted values for low-dimensional feature data and calculate feature residuals. Then, it combines the feature residuals with a dynamic threshold constructed based on the statistical features of the pressure signal to determine whether to trigger pulse transmission. This improves the problem that traditional pulse triggering methods mostly use fixed thresholds or only rely on the comparison of the amplitude of the original pressure signal. Due to the strong downhole pressure noise and drastic changes in operating conditions, this often results in frequent false triggering or missed triggering.
[0012] Furthermore, in the multi-source preprocessing module, the step of acquiring downhole multi-source time-series data includes:
[0013] Mud pressure data were collected at the first sampling rate;
[0014] The operating condition data is collected at a second sampling rate, and the operating condition data includes drill bit rotation speed, mud flow rate and torque;
[0015] Logging data was acquired at a third sampling rate, and the logging data included gamma ray intensity and resistivity.
[0016] Furthermore, in the multi-source preprocessing module, the step of constructing a unified state vector based on the preprocessed multi-source time-series data includes:
[0017] Time synchronization is performed on the collected multi-source time-series data to unify the timestamps and align data points with different sampling rates;
[0018] Perform linear normalization or Z-score standardization on the synchronized multi-source time-series data to eliminate dimensional differences;
[0019] The normalized multi-source time series data are concatenated according to the time index to form a unified state vector.
[0020] Furthermore, in the feature dimensionality reduction module, the steps of performing low-complexity feature extraction and data dimensionality reduction processing on the unified state vector include:
[0021] Sliding window statistics are performed on the unified state vector to extract the mean, variance, skewness and kurtosis within the window;
[0022] Calculate the first-order and second-order differences for the unified state vector;
[0023] The unified state vector is decomposed into two or three wavelet packets using Daubechies-4 wavelets to extract the energy of each frequency band.
[0024] Extract the temporal envelope from the unified state vector;
[0025] The sliding window statistics, difference results, wavelet packet frequency band energy, and time domain envelope are concatenated into low-dimensional feature data.
[0026] Furthermore, in the prediction triggering module, the step of real-time acquisition of downhole pressure signals includes:
[0027] The pressure signal output by the mud pressure sensor is continuously acquired at a sampling rate of not less than 200Hz;
[0028] Hardware low-pass filtering is applied to the acquired pressure signal;
[0029] The filtered pressure signal is stored in a circular buffer.
[0030] Furthermore, in the prediction triggering module, the step of constructing a dynamic threshold based on the statistical characteristics of the pressure signal includes:
[0031] The noise intensity, pressure variance, and pressure change rate are calculated based on the real-time collected pressure signals.
[0032] The noise intensity, pressure variance, and pressure change rate are normalized online using a sliding window to obtain normalized noise intensity, normalized pressure variance, and normalized pressure change rate.
[0033] The dynamic threshold is a weighted sum of the base threshold, normalized noise intensity, normalized pressure variance, and normalized pressure change rate.
[0034] Furthermore, in the prediction triggering module, the step of determining whether to trigger pulse transmission by combining feature residuals and dynamic thresholds includes:
[0035] Use a prediction model to generate predicted values for low-dimensional feature data;
[0036] The absolute value of the difference between the low-dimensional feature data and the predicted value is calculated as the original residual;
[0037] The original residuals are subjected to median filtering and exponential smoothing to obtain the effective residuals;
[0038] Compare the effective residual with the dynamic threshold. If the effective residual is greater than the dynamic threshold, then determine to trigger pulse transmission.
[0039] Furthermore, in the hierarchical compensation module, the step of performing hierarchical determination based on the feature residual size of the low-dimensional feature data includes:
[0040] Calculate the ratio of the effective residual to the dynamic threshold, and use it as the deviation ratio;
[0041] Compare the deviation ratio with the preset grading threshold;
[0042] If the deviation ratio is greater than 1 and less than or equal to the preset classification threshold, it is determined to be a first-level deviation.
[0043] If the deviation ratio is greater than the preset classification threshold, it is judged as a second-level deviation.
[0044] Furthermore, in the graded compensation module, the step of selecting the output standard pulse or performing pressure feedback compensation modulation on the pulse based on the graded determination result includes:
[0045] When the deviation is determined to be at the first level, a standard pulse is output.
[0046] When the deviation is determined to be level two, the pulse amplitude is adjusted according to the noise intensity in the statistical characteristics of the pressure signal, the pulse width is adjusted according to the pressure variance in the statistical characteristics of the pressure signal, the transmission interval is adjusted according to the pressure change rate in the statistical characteristics of the pressure signal, and the adjusted pulse is output.
[0047] Furthermore, in the closed-loop iteration module, the step of iteratively updating the prediction model parameters, dynamic threshold parameters, and impulse compensation parameters based on low-dimensional feature data and evaluation indicators includes:
[0048] Real-time downhole signals after pulse output are acquired, and pulse distortion, signal-to-noise ratio substitution index, and actual pulse rate are calculated as evaluation indicators.
[0049] The pulse compensation parameters are updated using pulse distortion, the dynamic threshold parameters are updated using signal-to-noise ratio as a substitute index, and the update amount is limited.
[0050] Trend-based prediction model parameters are updated based on feature residuals;
[0051] Boundary constraints are applied to the updated pulse compensation parameters, dynamic threshold parameters, and prediction model parameters.
[0052] The present invention has the following beneficial effects:
[0053] 1. In this invention, the prediction triggering module uses a prediction model to generate predicted values for low-dimensional feature data and calculates feature residuals. Then, it combines the feature residuals with a dynamic threshold constructed based on the statistical features of the pressure signal to determine whether to trigger pulse transmission. This improves the problem that traditional pulse triggering methods mostly use fixed thresholds or rely solely on the comparison of the amplitude of the original pressure signal. Due to the strong downhole pressure noise and drastic changes in operating conditions, this often results in frequent false triggering or missed triggering.
[0054] 2. In this invention, after the pulse transmission trigger takes effect, the graded compensation module performs graded judgment based on the size of the characteristic residual, and selects to output a standard pulse or perform pressure feedback compensation modulation according to the judgment result. When the noise intensity or pressure variance of the pressure signal exceeds the limit value determined based on the historical quantile, the reference pulse is fixedly output. This improves the problem that traditional pulse compensation methods mostly adopt a single compensation strategy or have no extreme working condition protection mechanism. Due to strong noise or abnormal pressure fluctuations, the compensation parameters may get out of control, resulting in aggravated pulse distortion or even complete signal loss.
[0055] 3. In this invention, evaluation indicators are constructed by collecting real-time downhole signals after pulse output through a closed-loop iterative module. Based on low-dimensional feature data and evaluation indicators, the prediction model parameters, dynamic threshold parameters, and pulse compensation parameters are iteratively updated. Boundary constraints are set during the iterative update process, thereby improving the problem that traditional parameter setting methods mostly use fixed parameters or open-loop adjustments. Due to the slow drift of the downhole environment over time and the lack of feedback constraints, the system performance deteriorates or parameters diverge after long-term operation. Attached Figure Description
[0056] Figure 1 This is a schematic diagram of the architecture of a downhole pulse generation control system based on pressure feedback compensation proposed in this invention.
[0057] Figure 2 This is a schematic flowchart of a downhole pulse generation control method based on pressure feedback compensation proposed in an embodiment of the present invention. Detailed Implementation
[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0059] Example 1: In the first embodiment of the present invention, the present invention provides a downhole pulse generation and control system based on pressure feedback compensation, such as... Figure 1As shown, the multi-source preprocessing module is used to collect downhole multi-source time-series data, preprocess the multi-source time-series data, and construct a unified state vector based on the preprocessed multi-source time-series data.
[0060] Furthermore, in the multi-source preprocessing module, the steps for acquiring downhole multi-source time-series data include:
[0061] Mud pressure data were collected at the first sampling rate;
[0062] Operating data was collected at the second sampling rate, including drill bit rotation speed, mud flow rate, and torque.
[0063] Logging data was acquired at the third sampling rate, and the logging data included gamma ray intensity and resistivity.
[0064] Specifically, in the multi-source preprocessing module, the acquisition of downhole multi-source time-series data is performed first; mud pressure data is then continuously acquired at the first sampling rate. Operating condition data is collected synchronously at the second sampling rate, including drill string rotation speed. Mud flow rate and torque Log data was acquired simultaneously at the third sampling rate, and the log data included gamma-ray intensity. and resistivity After data acquisition, the multi-source time-series data is time-synchronized. All data streams are resampled to a unified time base using linear interpolation or nearest neighbor interpolation, ensuring that each time step... The sensor data are aligned in time; then the synchronized data is normalized to eliminate dimensional differences between different physical quantities, for example, by using Z-score normalization for any data sequence. Its normalized value Calculated as ;in Let be the mean of this data series. Its standard deviation; the normalized mud pressure data, operating condition data, and well logging data are spliced together according to the time index to form a unified state vector. This unified state vector It serves as the sole input to the next-level feature dimensionality reduction module, and is used for subsequent low-complexity feature extraction and data dimensionality reduction processing.
[0065] The feature dimensionality reduction module is used to perform low-complexity feature extraction and data dimensionality reduction on a unified state vector. Feature extraction methods include one or more combinations of sliding window statistics, temporal difference, and low-order wavelet packet decomposition to generate low-dimensional feature data.
[0066] Furthermore, in the feature dimensionality reduction module, the steps for performing low-complexity feature extraction and data dimensionality reduction on the unified state vector include:
[0067] Sliding window statistics are performed on the unified state vector to extract the mean, variance, skewness and kurtosis within the window;
[0068] Calculate the first-order and second-order differences for the unified state vector;
[0069] The unified state vector is decomposed into two or three wavelet packets using Daubechies-4 wavelets to extract the energy of each frequency band.
[0070] Extract the temporal envelope from the unified state vector;
[0071] The sliding window statistics, difference results, wavelet packet frequency band energy, and time domain envelope are concatenated into low-dimensional feature data.
[0072] Specifically, in the feature dimensionality reduction module, the unified state vector is... Perform low-complexity feature extraction and data dimensionality reduction; input is a unified state vector. Provided by a multi-source preprocessing module, the components are normalized mud pressure, drill string speed, mud flow rate, torque, gamma ray intensity, and resistivity, respectively; firstly, sliding window statistics are performed, and the window length is set. It is 16 points, for Calculate the mean within the window for each dimension separately. ,variance skewness and kurtosis ;in for A certain dimension at time The value; then for Calculate the first difference and second-order difference The signal change rate characteristics were obtained; then, the Daubechies-4 wavelet was used to analyze the signal. Perform two-level wavelet packet decomposition to extract energy from each frequency band. ;in For the first The first of the sub-bands Each wavelet coefficient; then... Extract the time-domain envelope and calculate the magnitude of the analytic signal using Hilbert transform to obtain the envelope signal. The above sliding window statistical results, first-order and second-order difference results, wavelet packet frequency band energy, and time-domain envelope are concatenated in a fixed order to form low-dimensional feature data. This low-dimensional feature data The output is sent to the prediction trigger module, which is used for subsequent prediction model generation of predicted values and feature residual calculations to determine whether to trigger pulse transmission.
[0073] The prediction triggering module is used to acquire downhole pressure signals in real time, extract statistical features of the pressure signals, construct dynamic thresholds based on the statistical features of the pressure signals, generate predicted values for low-dimensional feature data using a prediction model, calculate feature residuals, and combine feature residuals with dynamic thresholds to determine whether to trigger pulse transmission.
[0074] Furthermore, in the prediction triggering module, the steps for real-time acquisition of downhole pressure signals include:
[0075] The pressure signal output by the mud pressure sensor is continuously acquired at a sampling rate of not less than 200Hz;
[0076] Hardware low-pass filtering is applied to the acquired pressure signal;
[0077] The filtered pressure signal is stored in a circular buffer.
[0078] Specifically, in the prediction triggering module, real-time acquisition of downhole pressure signals is achieved through the following steps: the analog signal output from the mud pressure sensor is processed by a signal conditioning circuit and then continuously acquired by an analog-to-digital converter at a sampling rate of not less than 200Hz to obtain the original pressure signal sequence. The raw pressure signal is then processed by a hardware low-pass filter, with the filter cutoff frequency set to five times the pulse fundamental frequency to suppress high-frequency noise interference, resulting in a filtered pressure signal. ; the filtered pressure signal The filtered pressure signal is stored in a circular buffer, with a length covering at least 10 pulse cycles, to ensure that subsequent processing can access sufficient historical data. Statistical features used for subsequent extraction of pressure signals, including noise intensity. Pressure variance and pressure change rate And as a construction of dynamic threshold The foundation.
[0079] Furthermore, in the prediction triggering module, the step of constructing a dynamic threshold based on the statistical characteristics of the pressure signal includes:
[0080] The noise intensity, pressure variance, and pressure change rate are calculated based on the real-time collected pressure signals.
[0081] The noise intensity, pressure variance, and pressure change rate are normalized online using a sliding window to obtain normalized noise intensity, normalized pressure variance, and normalized pressure change rate.
[0082] The dynamic threshold is a weighted sum of the base threshold, normalized noise intensity, normalized pressure variance, and normalized pressure change rate.
[0083] Specifically, in the prediction triggering module, the process of constructing a dynamic threshold based on the statistical characteristics of the pressure signal is as follows: the input is the pressure signal acquired in real time and then filtered. First, three statistical characteristics are calculated based on the pressure signal: noise intensity. ;in The moving average of the pressure signal; RMS represents the root mean square operation; pressure variance. That is, the variance of the pressure signal at the current moment; the rate of change of pressure. The above three statistical features are approximated using backward difference; then, online sliding window normalization is performed on the above three statistical features, maintaining a length of [missing information]. Historical queue, Take 100 to 200 points and calculate the moving average of each statistical characteristic. and sliding standard deviation The normalization formula is , , ;in Use a small positive number to prevent division by zero; finally, set the base threshold. The dynamic threshold is obtained by adding the weighted sum of the three normalized statistical features. Basic threshold The typical value range is 0.1 to 0.3, determined based on the 90th quantile of the characteristic residuals under normal downhole operating conditions, and the adjustable weighting coefficient. , and The typical value range is 0.2 to 0.5, where Take the maximum value to prioritize noise suppression. The minimum value is selected to avoid drastic fluctuations in the threshold caused by sudden changes in operating conditions; this dynamic threshold... The output is sent to the trigger determination stage and compared with the effective residual to determine whether to trigger pulse transmission.
[0084] Furthermore, in the prediction triggering module, the step of determining whether to trigger pulse transmission by combining feature residuals and dynamic thresholds includes:
[0085] Use a prediction model to generate predicted values for low-dimensional feature data;
[0086] The absolute value of the difference between the low-dimensional feature data and the predicted value is calculated as the original residual;
[0087] The original residuals are subjected to median filtering and exponential smoothing to obtain the effective residuals;
[0088] Compare the effective residual with the dynamic threshold. If the effective residual is greater than the dynamic threshold, then determine to trigger pulse transmission.
[0089] Specifically, in the prediction triggering module, the process of determining whether to trigger pulse transmission by combining feature residuals and dynamic thresholds is as follows: the input is the low-dimensional feature data output by the feature dimensionality reduction module. and the dynamic threshold constructed in the aforementioned steps First, use the prediction model to... Generate predicted values The prediction model uses a first-order autoregressive model. ;in and These are the model coefficients. The initial parameters of the model are the low-dimensional feature data from the previous time step. Take a value of 0.8 to 0.9. The value is set to 0.05~0.15, and subsequently updated in real time by the closed-loop iteration module; then the original residual is calculated. That is, the absolute value of the difference between the low-dimensional feature data and the predicted value; for the original residual Median filtering is performed, with a window radius of... Choosing 2 or 3 yields the filtered residual. Next, the filtered residuals are exponentially smoothed to obtain the effective residuals. ;in This is a smoothing coefficient, with a value ranging from 0.4 to 0.7. The effective residuals are from the previous time step; finally, the effective residuals are compared. With dynamic threshold ,like If the trigger pulse is sent, the trigger pulse will be sent; otherwise, the trigger pulse will not be sent. This trigger determination result is used to control the start of the graded compensation module. That is, the graded compensation module will only perform subsequent graded determination and pulse output operations after the trigger pulse is sent.
[0090] The graded compensation module is used to perform graded judgment based on the feature residual of low-dimensional feature data after the pulse transmission trigger takes effect. Based on the graded judgment result, it selects to output a standard pulse or perform pressure feedback compensation modulation on the pulse. When the noise intensity of the pressure signal exceeds the noise intensity limit value determined based on the historical quantile, or the pressure variance exceeds the pressure variance limit value determined based on the historical quantile, the reference pulse is fixedly output.
[0091] Furthermore, in the hierarchical compensation module, the step of performing hierarchical determination based on the feature residuals of the low-dimensional feature data includes:
[0092] Calculate the ratio of the effective residual to the dynamic threshold, and use it as the deviation ratio;
[0093] Compare the deviation ratio with the preset grading threshold;
[0094] If the deviation ratio is greater than 1 and less than or equal to the preset classification threshold, it is determined to be a first-level deviation.
[0095] If the deviation ratio is greater than the preset classification threshold, it is judged as a second-level deviation.
[0096] Specifically, in the hierarchical compensation module, the process of performing hierarchical determination based on the feature residual size of the low-dimensional feature data is as follows: the input is the effective residual calculated by the prediction trigger module. and dynamic threshold First, calculate the ratio of the effective residual to the dynamic threshold to obtain the deviation ratio. Then the deviation ratio With preset hierarchical threshold Comparison, preset hierarchical thresholds It is a constant greater than 1, with a typical value range of 1.5 to 2.5; if the deviation ratio satisfy If the deviation is greater than the first level, it is determined to be a first-level deviation; if the deviation is greater than the first level, it is determined to be a first-level deviation. satisfy If the deviation is not met, it is determined to be a second-level deviation. This classification result is used to control the strategy of subsequent pulse output: when the deviation is first-level, a standard basic pulse is output; when the deviation is second-level, pressure feedback compensation modulation is enabled to adjust the pulse waveform.
[0097] Furthermore, in the graded compensation module, the steps of selecting the output standard pulse or performing pressure feedback compensation modulation on the pulse based on the graded determination result include:
[0098] When the deviation is determined to be at the first level, a standard pulse is output.
[0099] When the deviation is determined to be level two, the pulse amplitude is adjusted according to the noise intensity in the statistical characteristics of the pressure signal, the pulse width is adjusted according to the pressure variance in the statistical characteristics of the pressure signal, the transmission interval is adjusted according to the pressure change rate in the statistical characteristics of the pressure signal, and the adjusted pulse is output.
[0100] Specifically, in the graded compensation module, the process of selecting the output standard pulse or performing pressure feedback compensation modulation on the pulse based on the graded judgment result is as follows: First, determine the noise intensity limit value N. max and the limit value of pressure variance σ max After the downhole tool is lowered to the working depth, pressure data is collected for the first 5 to 10 minutes, and the noise intensity sequence N is calculated. p (t) and pressure variance sequence Take the 95th percentile as N max and σ maxDuring real-time control, if the current noise intensity N p (t) is greater than N max Or current pressure variance Greater than σ max If the aforementioned extreme working condition is not triggered, a reference pulse will be forcibly output. This reference pulse has a fixed amplitude A0, width W0, and transmission interval T0, and subsequent pressure feedback compensation modulation will not be performed. The standard pulse and the reference pulse use the same basic parameters, where the amplitude A0 is 0.5~1.5MPa, the width W0 is 20~50ms, and the transmission interval T0 is 200~500ms, determined according to the downhole mud discharge and transmission rate requirements. Otherwise, if the aforementioned extreme working condition fallback condition is not triggered, i.e., the current noise intensity N is within the range of 0.5~1.5MPa, the standard pulse and the reference pulse use the same basic parameters, where the amplitude A0 is 0.5~1.5MPa, the width W0 is 20~50ms, and the transmission interval T0 is 200~500ms, determined according to the downhole mud discharge and transmission rate requirements. p (t)≤N max And the current pressure variance ≤σ max At that time, the input consists of the grading determination result and the statistical characteristics of the pressure signal, including noise intensity. Pressure variance and pressure change rate These features have already undergone online sliding window normalization in the previous steps, resulting in normalized noise intensity. Normalized pressure variance and normalized rate of change of pressure When the judgment result is a first-level deviation, a standard pulse is output, which has a fixed amplitude. ,width and transmission interval When the judgment result is a second-level deviation, pressure feedback compensation modulation is activated, based on the normalized noise intensity. Adjust the pulse amplitude to ;in This is the pulse amplitude compensation coefficient; based on the normalized pressure variance. Adjust the pulse width to ,in This is the pulse width compensation coefficient; based on the normalized pressure change rate. Adjust the sending interval to ;in For transmission interval compensation coefficient, pulse amplitude compensation coefficient Use a value of 0.2~0.6MPa for the pulse width compensation coefficient. The transmission interval compensation coefficient is set to 5-15ms. The pulse length is set to 50-150ms, determined based on the maximum permissible pulse distortion in downhole conditions; the adjusted pulse... The output is executed by the downhole pulse generator; the output pulse is used to transmit downhole data to the surface in the form of pressure fluctuations. At the same time, the actual output pulse parameters will be collected by the closed-loop iterative module to construct evaluation indicators and drive subsequent adaptive parameter updates.
[0101] The closed-loop iteration module is used to collect real-time downhole signals after pulse output to construct evaluation indicators. Based on low-dimensional feature data and evaluation indicators, iteratively update the prediction model parameters, dynamic threshold parameters and pulse compensation parameters. Boundary constraints are set during the iterative update process to complete the closed-loop control of the entire process.
[0102] Furthermore, in the closed-loop iterative module, the steps of iteratively updating the prediction model parameters, dynamic threshold parameters, and impulse compensation parameters based on low-dimensional feature data and evaluation indicators include:
[0103] Real-time downhole signals after pulse output are acquired, and pulse distortion, signal-to-noise ratio substitution index, and actual pulse rate are calculated as evaluation indicators.
[0104] The pulse compensation parameters are updated using pulse distortion, the dynamic threshold parameters are updated using signal-to-noise ratio as a substitute index, and the update amount is limited.
[0105] Trend-based prediction model parameters are updated based on feature residuals;
[0106] Boundary constraints are applied to the updated pulse compensation parameters, dynamic threshold parameters, and prediction model parameters.
[0107] Specifically, in the closed-loop iterative module, the process of iteratively updating the prediction model parameters, dynamic threshold parameters, and pulse compensation parameters based on low-dimensional feature data and evaluation indicators is as follows: The input consists of the actual pulse waveform parameters output by the graded compensation module and the downhole feedback signal. First, the downhole real-time signal after pulse output is collected, and three evaluation indicators are constructed: pulse distortion degree... ;in The pulse amplitude is the actual measured value at the ground or underground receiver. For the desired ideal pulse amplitude, Equal to the actual output pulse amplitude If the output standard pulse is equal to Signal-to-noise ratio (SNR) as a substitute indicator ;in The standard deviation of the pressure signal at the current moment; the actual pulse rate. ;in Time interval The actual number of pulses transmitted is then used; the pulse amplitude compensation coefficient is then updated using the pulse distortion degree. Pulse width compensation coefficient and transmission interval compensation coefficient The above parameters are used to adaptively adjust the pulse waveform parameters based on pressure statistical characteristics to improve anti-interference capability. The update rule is as follows: ;in The learning rate is the learning rate. Use values from 0.01 to 0.05, and limit the update amount: , The preset maximum single-step update magnitude; using signal-to-noise ratio as a substitute indicator. Update dynamic threshold parameters The update rule is as follows: Learning rate Using a value of 0.005~0.02, the update volume is also limited: , The preset amplitude limit is based on the original residual. The trend is analyzed, and the recursive least squares method is used to update the prediction model parameters. and This allows it to fit the changing pattern of the current residual sequence; finally, the updated pulse compensation parameters are adjusted. Dynamic threshold parameters and prediction model parameters Apply boundary constraints: The updated parameters are fed back to the feature reduction module, the prediction triggering module, and the hierarchical compensation module, respectively, to complete the closed-loop control of the entire process.
[0108] Example 2: In the second embodiment of the present invention, the present invention provides a downhole pulse generation control method based on pressure feedback compensation, such as... Figure 2 As shown, it includes the following steps:
[0109] Collect downhole multi-source time series data, preprocess the multi-source time series data, and construct a unified state vector based on the preprocessed multi-source time series data;
[0110] Low-complexity feature extraction and data dimensionality reduction are performed on the unified state vector. The feature extraction methods include sliding window statistics, temporal difference, and low-order wavelet packet decomposition to generate low-dimensional feature data.
[0111] Real-time acquisition of downhole pressure signals, extraction of statistical features of pressure signals, and construction of dynamic thresholds based on the statistical features of pressure signals; generation of predicted values for low-dimensional feature data using a prediction model, calculation of feature residuals, and determination of whether to trigger pulse transmission by combining feature residuals with dynamic thresholds;
[0112] After the pulse transmission trigger takes effect, a graded judgment is performed based on the magnitude of the feature residual of the low-dimensional feature data. Based on the graded judgment result, a standard pulse is output or the pulse is subjected to pressure feedback compensation modulation. When the noise intensity of the pressure signal exceeds the noise intensity limit value determined based on the historical quantile, or the pressure variance exceeds the pressure variance limit value determined based on the historical quantile, a fixed reference pulse is output.
[0113] Evaluation indices are constructed by collecting real-time downhole signals after pulse output. Based on low-dimensional feature data and evaluation indices, prediction model parameters, dynamic threshold parameters, and pulse compensation parameters are iteratively updated. Boundary constraints are set during the iterative update process to complete the closed-loop control of the entire process.
[0114] In oil drilling measurement-while-drilling (MWD) operations, downhole tools need to transmit geological data such as gamma rays and resistivity near the drill bit, as well as engineering parameters such as well inclination and azimuth, to the surface in real time via mud pulse generators. However, the downhole environment is subject to various strong noise interferences, including mud pump pressure fluctuations, drill string vibration, and formation fluid intrusion, resulting in low pressure signal-to-noise ratio and severe pulse distortion. Simultaneously, the pressure statistical characteristics vary drastically across different well sections and at different flow rates, making it highly susceptible to false triggering, missed triggering, or surface decoding failures when using fixed trigger thresholds and pulse parameters, severely impacting the reliability and real-time performance of data transmission. To address these issues, this invention provides a downhole pulse generation control method based on pressure feedback compensation, the process of which is as follows: Figure 2 As shown. The specific implementation process of this method is as follows:
[0115] First, multi-source time-series data from downhole are collected and preprocessed to construct a unified state vector. By fusing multi-dimensional information such as mud pressure, drill string speed, mud flow rate, torque, gamma rays, and resistivity, the correlation between working conditions and geological data can be effectively utilized to suppress the uncertainty of a single pressure signal, thereby improving the robustness of subsequent feature extraction.
[0116] Then, lightweight time-frequency feature extraction and dimensionality reduction processing such as sliding window statistics, time-domain difference and low-order wavelet packet decomposition are performed on the unified state vector to generate low-dimensional feature data. This measure greatly compresses the data dimension under the condition of limited computing power downhole, and retains the key information that best reflects the pressure fluctuation trend, laying the foundation for subsequent accurate prediction.
[0117] Next, downhole pressure signals are acquired in real time and their statistical features such as noise intensity, variance and rate of change are extracted to construct a dynamic threshold. At the same time, a prediction model is used to generate predicted values for low-dimensional feature data and calculate feature residuals. The feature residuals are compared with the dynamic threshold to determine whether to trigger pulse transmission. This method does not rely on a fixed pressure threshold, but adaptively adjusts the triggering criteria according to the current pressure statistical characteristics, thereby significantly reducing the false triggering rate and missed triggering rate in the background of strong noise.
[0118] After the trigger pulse is sent, a graded judgment is performed based on the size of the characteristic residual. Based on the judgment result, a standard pulse is output or pressure feedback compensation modulation is performed on the pulse amplitude, width and transmission interval. When the noise intensity or pressure variance exceeds the limit value determined based on the historical quantile, a reference pulse is forcibly output. Through graded control, energy is saved when the deviation is small and the anti-interference capability is enhanced when the deviation is large. At the same time, the fallback protection is activated under extreme noise to prevent the signal from being completely lost due to compensation failure.
[0119] Finally, real-time downhole signals after pulse output are collected to construct evaluation indicators such as pulse distortion, signal-to-noise ratio substitution index, and actual pulse rate. Based on these evaluation indicators and low-dimensional feature data, the prediction model parameters, dynamic threshold parameters, and pulse compensation parameters are iteratively updated. During the update process, the step size is limited and boundary constraints are applied to the parameters, thereby forming a closed-loop control throughout the entire process, enabling the system to operate stably for a long time and always maintain a high recognition rate and a low bit error rate.
[0120] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A downhole pulse generation and control system based on pressure feedback compensation, characterized in that, include: The multi-source preprocessing module is used to collect downhole multi-source time-series data, preprocess the multi-source time-series data, and construct a unified state vector based on the preprocessed multi-source time-series data. The feature dimensionality reduction module is used to perform low-complexity feature extraction and data dimensionality reduction on a unified state vector. The feature extraction methods include sliding window statistics, temporal difference, and low-order wavelet packet decomposition to generate low-dimensional feature data. The prediction triggering module is used to acquire downhole pressure signals in real time, extract statistical features of the pressure signals, and construct dynamic thresholds based on the statistical features of the pressure signals. The prediction model is used to generate predicted values for low-dimensional feature data and calculate feature residuals. The feature residuals are combined with dynamic thresholds to determine whether to trigger pulse transmission. The graded compensation module is used to perform graded judgment based on the feature residual of low-dimensional feature data after the pulse transmission trigger takes effect. Based on the graded judgment result, it selects to output a standard pulse or perform pressure feedback compensation modulation on the pulse. When the noise intensity of the pressure signal exceeds the noise intensity limit value determined based on the historical quantile, or the pressure variance exceeds the pressure variance limit value determined based on the historical quantile, the reference pulse is fixedly output. The closed-loop iteration module is used to collect real-time downhole signals after pulse output to construct evaluation indicators. Based on low-dimensional feature data and evaluation indicators, iteratively update the prediction model parameters, dynamic threshold parameters, and pulse compensation parameters. Boundary constraints are set during the iterative update process to complete the closed-loop control of the entire process.
2. The downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the multi-source preprocessing module, the steps for acquiring downhole multi-source time-series data include: Mud pressure data were collected at the first sampling rate; The operating condition data is collected at a second sampling rate, and the operating condition data includes drill bit rotation speed, mud flow rate and torque; Logging data was acquired at a third sampling rate, and the logging data included gamma ray intensity and resistivity.
3. The downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the multi-source preprocessing module, the step of constructing a unified state vector based on the preprocessed multi-source time-series data includes: Time synchronization is performed on the collected multi-source time-series data to unify the timestamps and align data points with different sampling rates; Perform linear normalization or Z-score standardization on the synchronized multi-source time-series data to eliminate dimensional differences; The normalized multi-source time series data are concatenated according to the time index to form a unified state vector.
4. The downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the feature dimensionality reduction module, the steps of performing low-complexity feature extraction and data dimensionality reduction on the unified state vector include: Sliding window statistics are performed on the unified state vector to extract the mean, variance, skewness and kurtosis within the window; Calculate the first-order and second-order differences for the unified state vector; The unified state vector is decomposed into two or three wavelet packets using Daubechies-4 wavelets to extract the energy of each frequency band. Extract the temporal envelope from the unified state vector; The sliding window statistics, difference results, wavelet packet frequency band energy, and time domain envelope are concatenated into low-dimensional feature data.
5. A downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the prediction triggering module, the step of real-time acquisition of downhole pressure signals includes: The pressure signal output by the mud pressure sensor is continuously acquired at a sampling rate of not less than 200Hz; Hardware low-pass filtering is applied to the acquired pressure signal; The filtered pressure signal is stored in a circular buffer.
6. The downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the prediction triggering module, the step of constructing a dynamic threshold based on the statistical characteristics of the pressure signal includes: The noise intensity, pressure variance, and pressure change rate are calculated based on the real-time collected pressure signals. The noise intensity, pressure variance, and pressure change rate are normalized online using a sliding window to obtain normalized noise intensity, normalized pressure variance, and normalized pressure change rate. The dynamic threshold is a weighted sum of the base threshold, normalized noise intensity, normalized pressure variance, and normalized pressure change rate.
7. A downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the prediction triggering module, the step of determining whether to trigger pulse transmission by combining feature residuals and dynamic thresholds includes: Use a prediction model to generate predicted values for low-dimensional feature data; The absolute value of the difference between the low-dimensional feature data and the predicted value is calculated as the original residual; The original residuals are subjected to median filtering and exponential smoothing to obtain the effective residuals; Compare the effective residual with the dynamic threshold. If the effective residual is greater than the dynamic threshold, then determine to trigger pulse transmission.
8. A downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the hierarchical compensation module, the step of performing hierarchical determination based on the feature residual size of the low-dimensional feature data includes: Calculate the ratio of the effective residual to the dynamic threshold, and use it as the deviation ratio; Compare the deviation ratio with the preset grading threshold; If the deviation ratio is greater than 1 and less than or equal to the preset classification threshold, it is determined to be a first-level deviation. If the deviation ratio is greater than the preset classification threshold, it is judged as a second-level deviation.
9. A downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the graded compensation module, the step of selecting the output standard pulse or performing pressure feedback compensation modulation on the pulse based on the graded determination result includes: When the deviation is determined to be at the first level, a standard pulse is output. When the deviation is determined to be level two, the pulse amplitude is adjusted according to the noise intensity in the statistical characteristics of the pressure signal, the pulse width is adjusted according to the pressure variance in the statistical characteristics of the pressure signal, the transmission interval is adjusted according to the pressure change rate in the statistical characteristics of the pressure signal, and the adjusted pulse is output.
10. A downhole pulse generation and control system based on pressure feedback compensation according to claim 1, characterized in that, In the closed-loop iteration module, the steps of iteratively updating the prediction model parameters, dynamic threshold parameters, and impulse compensation parameters based on low-dimensional feature data and evaluation indicators include: Real-time downhole signals after pulse output are acquired, and pulse distortion, signal-to-noise ratio substitution index, and actual pulse rate are calculated as evaluation indicators. The pulse compensation parameters are updated using pulse distortion, the dynamic threshold parameters are updated using signal-to-noise ratio as a substitute index, and the update amount is limited. Trend-based prediction model parameters are updated based on feature residuals; Boundary constraints are applied to the updated pulse compensation parameters, dynamic threshold parameters, and prediction model parameters.