Artificial intelligence-based survey operation anomaly monitoring system and method

Abstract

The application discloses a survey operation abnormality monitoring system and method based on artificial intelligence, and relates to the technical field of survey operation monitoring.The method comprises the following steps: demarcating a cycle boundary according to a drill rod connection and disconnection event, extracting initial period behavior data, generating multi-dimensional reference features and initial correlation coefficients between dimensions, and taking self-consistency verification qualification as an effective condition; setting the first data segment as a dynamic reference, updating the dynamic reference when the deviation of the subsequent segment from the multi-dimensional reference features and the dynamic reference is in a steady state interval, otherwise freezing the update authority; monitoring the drift trend of each dimension, performing stratum trend stripping when a single dimension drifts, and calculating the deviation degree of real-time correlation coefficients when all dimensions drift; triggering a change mode decision when the deviation degree continuously increases and has no linear dependence on the depth of the footage, and determining a drill rod misplacement abnormality when the change rate jumps in steps and the step time point is synchronized with the turning point of each dimension trend.

Description

Technical Field

[0001] This invention relates to the field of exploration operation monitoring technology, specifically to an artificial intelligence-based exploration operation anomaly monitoring system and method. Background Technology

[0002] In engineering exploration drilling operations, drill pipe assemblies are composed of multiple short drill pipe sections connected by threads. Individual drill pipes are short and require frequent connection and disconnection, and the drilling depth depends on the cumulative calculation of the number of drill pipe sections. The mechanical connection status of the drill pipe assembly within the hole directly affects drilling safety and core sampling quality.

[0003] Existing monitoring methods typically rely on sensors to collect physical quantities such as vibration and pressure to trigger threshold exceeding alarms. These methods can only detect significant anomalies that have already occurred, lacking the ability to detect progressive early failures such as slow loosening of joint threads or hidden relative slippage between drill pipes. Methods based on visual recognition to directly detect surface defects on drill pipes or abnormal joint appearances have a high false alarm rate under conditions of mud and rock powder cover at the exploration site, and the internal meshing state of joints cannot be directly judged by appearance. Changes in torque, drilling speed, or footage curves during drilling may originate from formation changes or alterations in the internal connection state of the drill pipe assembly; conventional methods struggle to reliably distinguish between these two factors early on based solely on changes in the amplitude of a single parameter. Furthermore, current technologies have not yet translated the fundamental technological constraint of maintaining the rigidity of the drill pipe assembly within the same run into a calculable monitoring logic for utilization. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based system and method for monitoring anomalies in exploration operations, in order to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based method for monitoring anomalies in exploration operations, the method comprising: S10. Based on the identification results of drill pipe unloading events, mark the return boundary, extract the drill pipe assembly behavior data in the initial period of the return, generate multi-dimensional benchmark features and initial correlation coefficients between dimensions to characterize the overall dynamic characteristics of the drill pipe assembly, and take the self-consistency verification of the initial period data as the effective condition for the multi-dimensional benchmark features and initial correlation coefficients. S20. Set the behavior feature of the first data segment in the cycle as the dynamic benchmark and start the dynamic benchmark refresh cycle. When the deviations of subsequent data segments from the multidimensional benchmark features and from the current dynamic benchmark are both in the steady state range, include the corresponding data segment data and update the dynamic benchmark. When any deviation enters the non-steady state range, immediately freeze the dynamic benchmark update permission. S30. For each data segment, synchronously monitor the drift trend of each dimension feature relative to the current dynamic benchmark; if only a single dimension shows unidirectional drift, perform stratigraphic trend stripping and correct the feature value of the current data segment before re-executing drift monitoring; if all dimensions show unidirectional drift, calculate the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient. S40. When the degree of offset continues to increase, and the increasing trend of the degree of offset has no linear dependence on the change of the depth of advance per pass, the change pattern judgment of the degree of offset is triggered. S50. Monitor the rate of change of the offset. When the rate of change changes by a step jump and the step time point is synchronized with the trend inflection point of the drift curve of each dimension, it is determined that an abnormality of drill pipe misalignment has occurred.

[0006] According to the above scheme, step S10 includes: S11. Obtain the identification results of drill pipe unloading events. Take the pure drilling interval without unloading events as a single cycle, take the time of the last unloading completion in the cycle as the cycle start boundary, and take the time of the first unloading start in the cycle as the cycle end boundary. S12. From the first steady-state drilling period after the initial boundary of the cycle, extract the continuous data with statistical stability as the initial period data. The steady-state drilling period refers to the continuous period in which the fluctuation range of drill pipe rotation speed and drilling pressure are both within the steady-state judgment range. S13. Extract the rod rotation smoothness characteristics, axial feed smoothness characteristics, and rod vibration frequency domain characteristics of the drill rod assembly from the initial time period data to construct multi-dimensional benchmark characteristics; S14. Calculate the correlation coefficient between any two dimensions in the multidimensional baseline feature, and use it as the initial correlation coefficient between dimensions. S15. Divide the initial time period data into several verification data segments according to time, and calculate the multidimensional feature values ​​in each verification data segment. When the deviations between the multidimensional feature values ​​of each verification data segment are all within the same stable distribution range, the multidimensional benchmark feature and the initial correlation coefficient are deemed to be effective. If the verification fails, slide the data capture window backward, re-extract the initial time period data, and complete the feature construction and consistency verification until the verification is successful.

[0007] According to the above scheme, step S20 includes: S21. After the initial boundary of the cycle, the multidimensional feature values ​​and interdimensional correlation coefficients extracted in the first data segment are set as dynamic benchmarks. The first data segment is a continuous data segment of the same length as the initial data segment, which is truncated from the end time of the initial time segment data. S22. Start the dynamic benchmark refresh loop, take subsequent data segments in sequence, and calculate the first deviation between the multidimensional feature value of the current data segment and the multidimensional benchmark feature, and the second deviation between the current data segment and the dynamic benchmark. S23. When the first and second deviations of a continuous data segment are both in the steady-state range, the data of the continuous data segment is included in the dynamic benchmark sample set, and the dynamic benchmark is updated with the statistical center of the dynamic benchmark sample set. S24. When the first or second deviation of any data segment enters the non-steady-state region, immediately freeze the dynamic benchmark update permission and suspend the expansion of the dynamic benchmark sample set and the update of the dynamic benchmark. S25. Continuously monitor the first and second deviations of subsequent data segments. When a data segment that has been continuously in the steady-state range regains the statistical stability required for benchmark updates, unfreeze it and re-include the data segment that has been continuously in the steady-state range into the dynamic benchmark sample set. Update the dynamic benchmark again with the statistical center of the dynamic benchmark sample set.

[0008] According to the above scheme, step S30 includes: S31. For each data segment, extract the characteristics of rod rotation smoothness, axial feed smoothness, and rod vibration frequency domain, respectively, as the feature values ​​of each dimension of the current data segment. S32. Calculate the deviation of each dimension feature value of the current data segment from the corresponding dimension feature value in the dynamic benchmark, and determine whether the deviation of each dimension is in the non-steady-state interval and continues to accumulate in the same direction. S33. When only one dimension of the deviation is in the unsteady range and drifts in one direction, while the deviations of other dimensions are still in the steady range, the stratigraphic trend component that changes monotonically with the footage depth is extracted from the current data segment. The extracted data is used to correct the characteristic value of the current data segment, and the deviation of each dimension is recalculated with the corrected characteristic value. Then, it is determined again whether the deviation of each dimension is in the unsteady range and continues to accumulate in the same direction. S34. When the deviations in all dimensions are in the unsteady range and all drift in one direction, calculate the first real-time correlation coefficient between the rotational smoothness feature and the axial feed smoothness feature in the current data segment, the second real-time correlation coefficient between the rotational smoothness feature and the vibration frequency domain feature of the rod, and the third real-time correlation coefficient between the axial feed smoothness feature and the vibration frequency domain feature of the rod. S35. Compare the first real-time correlation coefficient, the second real-time correlation coefficient, and the third real-time correlation coefficient with the corresponding correlation coefficients in the initial correlation coefficients, calculate the deviation of each correlation coefficient, and use the statistical mean of the deviation of each correlation coefficient as the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient.

[0009] According to the above scheme, step S40 includes: S41. Obtain the degree of offset calculated from continuous data segments with statistical stability for trend determination, and form a time series of offset. S42. Determine whether the time series of offset shows a continuous increasing trend. A continuous increasing trend means that within the corresponding continuous data segment, the offset accumulates and increases in a single direction, and the accumulated increase exceeds the upper limit of the steady-state interval of the offset. S43. When the offset time series shows a continuous increasing trend, extract the per-pass depth change series corresponding to the continuous increasing trend, and determine whether there is a linear dependency between the increasing trend of offset and the per-pass depth change. S44. When there is no linear dependence between the increasing trend of the offset degree and the change in the depth of each pass, the offset degree change pattern judgment is triggered, and the monitoring of the offset degree change rate is initiated. S45. When there is a linear dependence between the increasing trend of the offset degree and the change of the depth of each pass, the change mode decision is not triggered. The drift trend of each dimension feature relative to the current dynamic benchmark continues to be monitored synchronously for subsequent data segments, and the offset degree of the real-time correlation coefficient between dimensions relative to the initial correlation coefficient is calculated.

[0010] According to the above scheme, step S50 includes: S51. Obtain the offset degree of continuous data segments after the trigger change mode decision, and form an offset degree time series; calculate the first difference of the offset degree between adjacent data segments in the offset degree time series, and use the first difference to characterize the rate of change of the offset degree, and form a rate of change series. S52. Monitor the rate of change sequence and determine whether there is a step jump in the rate of change from the previous steady state level to the next higher steady state level. A step jump means that the increase in the rate of change between adjacent data segments exceeds the upper limit of the steady state range of the rate of change, and the rate of change after the jump remains at the new steady state level. S53. When a step jump is detected, record the time point corresponding to the starting data segment of the jump as the step time point. S54. Extract the drift curves of the drift of each dimension feature value relative to the current dynamic benchmark over time, and identify the trend turning points where the drift direction or drift slope changes on the drift curves of each dimension. S55. Determine whether the time deviation between the step time point and the trend turning point of the drift curve in each dimension is within the synchronous judgment interval. When the trend turning point of all dimensions is synchronized with the step time point, it is determined that a drill pipe misalignment abnormality has occurred.

[0011] An AI-based exploration operation anomaly monitoring system, which includes: a baseline construction module, a dynamic refresh module, a coupled offset module, and an anomaly judgment module; The benchmark construction module is used to calibrate the return boundary based on the identification results of drill pipe unloading events, extract drill pipe assembly behavior data in the initial period of the return, and generate multi-dimensional benchmark features and initial correlation coefficients between dimensions to characterize the overall dynamic characteristics of the drill pipe assembly. The self-consistency verification of the initial period data is used as the effective condition for the multi-dimensional benchmark features and initial correlation coefficients. The dynamic refresh module is used to set the behavioral characteristics of the first data segment in a cycle as the dynamic benchmark and start the dynamic benchmark refresh cycle. When the deviations of subsequent data segments from the multidimensional benchmark features and from the current dynamic benchmark are both in the steady-state range, the corresponding data segments are included and the dynamic benchmark is updated. When any deviation enters the non-steady-state range, the update permission of the dynamic benchmark is immediately frozen. The coupling offset module is used to synchronously monitor the drift trend of each dimension feature relative to the current dynamic benchmark for each data segment. If only a single dimension shows unidirectional drift, the stratigraphic trend stripping is performed and the feature value of the current data segment is corrected before the drift monitoring is re-executed. If all dimensions show unidirectional drift, the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient is calculated. The anomaly detection module is used to trigger a change mode detection when the degree of offset continues to increase and its increasing trend has no linear dependence on the change in the depth of each pass, and to monitor the rate of change of the degree of offset; when the rate of change changes by a step jump and the step time point is synchronized with the trend inflection point of the drift curve of each dimension, it is determined that a drill pipe misalignment anomaly has occurred.

[0012] According to the above scheme, the dynamic refresh module includes a baseline update unit and a baseline freeze unit; The benchmark update unit is used to include the data of the continuous data segment into the dynamic benchmark sample set when the first deviation of the continuous data segment from the multidimensional benchmark features and the second deviation from the current dynamic benchmark are both in the steady state range, and to update the dynamic benchmark with the statistical center of the dynamic benchmark sample set. The benchmark freeze unit is used to immediately freeze the dynamic benchmark update permission when the first or second deviation of any data segment enters the non-steady-state interval, suspending the expansion of the dynamic benchmark sample set and the update of the dynamic benchmark; when the data segment that has been continuously in the steady-state interval regains the statistical stability required for benchmark update, the freeze is lifted, and the data segment that has been continuously in the steady-state interval is re-included into the dynamic benchmark sample set, and the dynamic benchmark is updated again with the statistical center of the dynamic benchmark sample set.

[0013] According to the above scheme, the coupling offset module includes a drift monitoring unit and an offset quantization unit; The drift monitoring unit is used to calculate the deviation of each dimension feature value of each data segment from the corresponding dimension feature value in the dynamic benchmark. When only the deviation of a single dimension is in the non-steady-state range and drifts in one direction, while the deviations of other dimensions are still in the steady-state range, the stratigraphic trend component that changes monotonically with the footage depth is extracted from the current data segment. The extracted data is used to correct the feature value of the current data segment, and the deviation of each dimension is recalculated with the corrected feature value. Then, it is determined again whether the deviation of each dimension is in the non-steady-state range and continues to accumulate in the same direction. The offset quantization unit is used to calculate the real-time correlation coefficients between the features of each dimension in the current data segment when the deviations of all dimensions are in the non-steady-state range and all are unidirectional drifts. It then compares each real-time correlation coefficient with the corresponding correlation coefficient in the initial correlation coefficient and uses the statistical mean of the deviation of each correlation coefficient as the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient.

[0014] According to the above scheme, the anomaly detection module includes a trend verification unit and a step detection unit; The trend verification unit is used to obtain the offset degree calculated from continuous data segments with statistical stability for trend determination, forming an offset degree time series; determine whether the offset degree time series shows a continuous increasing trend; when a continuous increasing trend is shown, determine whether there is a linear dependency between the increasing trend of the offset degree and the change in the depth of each pass; when there is no linear dependency, trigger the change mode judgment and start the change rate monitoring; when there is a linear dependency, do not trigger the change mode judgment, and continue to synchronously monitor the drift trend of each dimension feature relative to the current dynamic benchmark for subsequent data segments and calculate the offset degree of the real-time correlation coefficient between dimensions relative to the initial correlation coefficient; The step detection unit is used to obtain the degree of offset of continuous data segments after the trigger change mode decision, calculate the first-order difference of the offset between adjacent data segments to characterize the rate of change; determine whether there is a step jump in the rate of change from the previous steady-state level to the next higher steady-state level; when a step jump is detected, record the time point corresponding to the starting data segment of the jump as the step time point; extract the drift curve of the drift of each dimension feature value of each data segment relative to the current dynamic benchmark over time, and identify the trend inflection points on the drift curves of each dimension; when the trend inflection points of all dimensions are synchronized with the step time point, it is determined that a drill pipe misalignment anomaly has occurred.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention establishes a multidimensional dynamic benchmark and initial dimensional correlation coefficient, distinguishes between single-dimensional stratigraphic drift and full-dimensional equipment coupling drift, removes stratigraphic trend components to eliminate environmental interference, and significantly reduces the misjudgment rate of anomaly monitoring; 2. This invention sets up a dual mechanism of dynamic benchmark refresh and freezing, which adaptively updates the benchmark based on the steady-state of segment deviation, ensuring that the benchmark always fits the real-time drilling conditions and improving the stability of the monitoring system under complex formations. 3. This invention captures early weak misalignment features by combining correlation offset timing, rate of change step, and multi-dimensional drift timing synchronization, thus overcoming the limitation of traditional threshold methods that can only identify drastic changes and achieving early warning of drill pipe misalignment. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the artificial intelligence-based anomaly monitoring method for exploration operations according to the present invention. Figure 2 This is a schematic diagram of the structure of the artificial intelligence-based exploration operation anomaly monitoring system of the present invention. Detailed Implementation

[0017] The technical solutions of 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.

[0018] Example: Figures 1-2 As shown, the present invention provides a technical solution: an artificial intelligence-based method for monitoring anomalies in exploration operations. The method includes: S10. Based on the identification results of drill pipe unloading events, mark the return boundary, extract the drill pipe assembly behavior data in the initial period of the return, generate multi-dimensional benchmark features and initial correlation coefficients between dimensions to characterize the overall dynamic characteristics of the drill pipe assembly, and take the self-consistency verification of the initial period data as the effective condition for the multi-dimensional benchmark features and initial correlation coefficients. S20. Set the behavior feature of the first data segment in the cycle as the dynamic benchmark and start the dynamic benchmark refresh cycle. When the deviations of subsequent data segments from the multidimensional benchmark features and from the current dynamic benchmark are both in the steady state range, include the corresponding data segment data and update the dynamic benchmark. When any deviation enters the non-steady state range, immediately freeze the dynamic benchmark update permission. S30. For each data segment, synchronously monitor the drift trend of each dimension feature relative to the current dynamic benchmark; if only a single dimension shows unidirectional drift, perform stratigraphic trend stripping and correct the feature value of the current data segment before re-executing drift monitoring; if all dimensions show unidirectional drift, calculate the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient. S40. When the degree of offset continues to increase, and the increasing trend of the degree of offset has no linear dependence on the change of the depth of advance per pass, the change pattern judgment of the degree of offset is triggered. S50. Monitor the rate of change of the offset. When the rate of change changes by a step jump and the step time point is synchronized with the trend inflection point of the drift curve of each dimension, it is determined that an abnormality of drill pipe misalignment has occurred.

[0019] Specifically, step S10 includes: S11. Obtain the identification results of drill pipe unloading events. Take the pure drilling interval without unloading events as a single cycle, take the time of the last unloading completion in the cycle as the cycle start boundary, and take the time of the first unloading start in the cycle as the cycle end boundary. S12. From the first steady-state drilling period after the initial boundary of the cycle, extract the continuous data with statistical stability as the initial period data. The steady-state drilling period refers to the continuous period in which the fluctuation range of drill pipe rotation speed and drilling pressure are both within the steady-state judgment range. S13. Extract the rod rotation smoothness characteristics, axial feed smoothness characteristics, and rod vibration frequency domain characteristics of the drill rod assembly from the initial time period data to construct multi-dimensional benchmark characteristics; S14. Calculate the correlation coefficient between any two dimensions in the multidimensional baseline feature, and use it as the initial correlation coefficient between dimensions. S15. Divide the initial time period data into several verification data segments according to time, and calculate the multidimensional feature values ​​in each verification data segment. When the deviations between the multidimensional feature values ​​of each verification data segment are all within the same stable distribution range, the multidimensional benchmark feature and the initial correlation coefficient are deemed to be effective. If the verification fails, slide the data capture window backward, re-extract the initial time period data, and complete the feature construction and consistency verification until the verification is successful.

[0020] Specifically, in this embodiment, an AI recognition model is used to identify drill pipe unloading events. The AI ​​recognition model is a machine learning model trained on labeled samples. The training samples are drilling site images labeled with two types of events: unloading completion and removal start. The AI ​​recognition model takes real-time video frames from the drilling site as input and outputs event category labels and corresponding confidence scores. When the confidence score for unloading completion in three consecutive frames exceeds 0.85, an unloading completion event is triggered, and the timestamp of the first frame in the three consecutive frames is taken as the unloading completion time. When the confidence score for removal start in three consecutive frames exceeds 0.85, a removal start event is triggered. The data is recorded, and the timestamp corresponding to the first frame of the three consecutive frames is taken as the unloading start time. In this embodiment, the unloading completion event is triggered at 37 minutes and 20 seconds, and the unloading start event is triggered at 118 minutes and 45 seconds. The 81 minutes and 25 seconds between the two is the pure drilling interval of this cycle, which is used to mark the start and end boundaries of the cycle. The initial time period data is extracted from the first steady-state drilling period after the start boundary of the cycle. A sliding window scan is performed on the drilling data after the start of the cycle. The window length is 60 seconds and the step size is 10 seconds. The coefficient of variation of the drill pipe rotation speed and the coefficient of variation of the drilling pressure in each window are calculated. The coefficient of variation is defined as the ratio of the standard deviation of the feature to the mean of the feature within the window, used to quantify the degree of data fluctuation. When both coefficients of variation are below 0.15, the window is determined to be a steady-state drilling period. In this embodiment, 0.15 is a fixed threshold for steady-state screening determined by statistical analysis of the coefficients of variation of manually calibrated steady-state periods in historical drilling data, taking the 95th percentile of the distribution of coefficients of variation of historical steady-state periods. This is only an example and is not a limitation. In this embodiment, the window from 39 minutes and 10 seconds to 40 minutes and 10 seconds meets the conditions, with a rotational speed coefficient of variation of 0.11 and a drilling pressure coefficient of variation of [missing value]. 0.09, the continuous data within this window is extracted as the initial time period data, with a duration of 60 seconds, a sampling rate of 10Hz, and a total of 600 sampling points; multi-dimensional benchmark features are extracted from the initial time period data; the rod rotation smoothness feature is taken as the standard deviation of the rotation speed time series, which is calculated to be 2.3 rpm; the axial feed smoothness feature is taken as the standard deviation of the first difference of the feed displacement, which is calculated to be 0.8 mm / s; the rod vibration frequency domain feature is taken as the energy proportion of the vibration acceleration signal in the 50-200Hz frequency band, which is calculated to be 0.62; in this embodiment, the energy proportion is calculated by fast Fourier transform, and the formula is: E ratio =Σ(f∈[50,200])|X(f)| 2 / Σ(f∈[0,fs / 2])|X(f)| 2 Where X(f) is the Fourier transform result of the vibration acceleration signal, f is the frequency, fs is the sampling rate (in this embodiment, fs = 10Hz), Σ(f∈[50,200]) represents the summation of energy in the 50-200Hz frequency band, and Σ(f∈[0,fs / 2]) represents the summation of energy in all frequency bands.ratio The frequency band energy proportion is used to construct a multi-dimensional baseline feature vector [2.3, 0.8, 0.62]. In this embodiment, the Pearson correlation coefficient is used to calculate the correlation coefficient r12 between the rod rotation smoothness feature and the axial feed smoothness feature, which is 0.73; the correlation coefficient r13 between the rod rotation smoothness feature and the rod vibration frequency domain feature is -0.41; and the correlation coefficient r23 between the axial feed smoothness feature and the rod vibration frequency domain feature is -0.28. These three coefficients constitute the initial correlation coefficient vector [0.73, -0.41, -0.28]. The formula for calculating the Pearson correlation coefficient is: r=Σ[(xi-X)(yi-Y)] / [Σ(xi-X)]. 2 ×Σ(yi-Y) 2 ] 1 / 2 Where xi and yi are the i-th sampled values ​​of the two dimensions, X and Y are the sampled mean values ​​of the two dimensions, Σ represents the summation over all sampled points, and r ranges from [-1, 1], where r approaching 1 represents a strong positive correlation, approaching -1 represents a strong negative correlation, and approaching 0 represents no correlation. The initial 60-second data period is divided into 5 verification data segments, each segment lasting 12 seconds, with a total of 120 sampled points. The three-dimensional feature values ​​within each verification segment are calculated as follows: the feature vector for segment 1 is [2.5, 0.7, 0.59], for segment 2 it is [2.1, 0.9, 0.64], for segment 3 it is [2.32, 0.78, 0.63], and for segment 4 it is [2.2, 0.7, 0.60]. Segment 5 is [2.3, 0.8, 0.61]. The coefficient of variation of each dimension of the feature values ​​of the five segments is calculated. The coefficient of variation of rotation smoothness is 0.07, the coefficient of variation of feed smoothness is 0.11, and the coefficient of variation of vibration frequency domain feature is 0.03. All of these are less than the consistency judgment threshold of 0.15. In this embodiment, 0.15 is the initial benchmark self-consistency judgment fixed threshold determined by statistically analyzing the coefficients of variation of the multidimensional feature values ​​of each verification segment in the initial time period data of multiple sets of normal drilling rounds. The 90th percentile value of the coefficient of variation distribution under normal working conditions is taken. All dimensions are lower than this value, which means that the initial data has no abnormal fluctuations and is statistically stable. The self-consistency verification is qualified, and the initial correlation coefficient vector is effective. This is only an example and is not a limitation.

[0021] Specifically, step S20 includes: S21. After the initial boundary of the cycle, the multidimensional feature values ​​and interdimensional correlation coefficients extracted in the first data segment are set as dynamic benchmarks. The first data segment is a continuous data segment of the same length as the initial data segment, which is truncated from the end time of the initial time segment data. S22. Start the dynamic benchmark refresh loop, take subsequent data segments in sequence, and calculate the first deviation between the multidimensional feature value of the current data segment and the multidimensional benchmark feature, and the second deviation between the current data segment and the dynamic benchmark. S23. When the first and second deviations of a continuous data segment are both in the steady-state range, the data of the continuous data segment is included in the dynamic benchmark sample set, and the dynamic benchmark is updated with the statistical center of the dynamic benchmark sample set. S24. When the first or second deviation of any data segment enters the non-steady-state region, immediately freeze the dynamic benchmark update permission and suspend the expansion of the dynamic benchmark sample set and the update of the dynamic benchmark. S25. Continuously monitor the first and second deviations of subsequent data segments. When a data segment that has been continuously in the steady-state range regains the statistical stability required for benchmark updates, unfreeze it and re-include the data segment that has been continuously in the steady-state range into the dynamic benchmark sample set. Update the dynamic benchmark again with the statistical center of the dynamic benchmark sample set.

[0022] Specifically, starting from the end of the initial data period, a continuous data segment of the same length as the initial data period is extracted, i.e., 60 seconds of data from 40 minutes and 10 seconds to 41 minutes and 10 seconds. Its multidimensional feature values ​​are extracted as [2.4, 0.7, 0.63], and the inter-dimensional correlation coefficients are [0.71, -0.43, -0.26]. This set of values ​​is set as the dynamic baseline. Subsequent data segments are then taken sequentially, each segment lasting 60 seconds. The first deviation between the current segment and the multidimensional baseline features, and the second deviation from the dynamic baseline, are calculated. The first deviation is the Euclidean distance between the current segment's feature values ​​and the multidimensional baseline feature vector [2.3, 0.8, 0.62]. The second deviation is the Euclidean distance between the current segment's feature values ​​and the dynamic baseline vector. The Euclidean distance calculation formula is: d = [(x1 - y1)]. 2 +(x²-y²) 2 +(x³-y³) 2 ] 1 / 2Where x1, x2, and x3 are the three-dimensional feature values ​​of the current segment, and y1, y2, and y3 are the three-dimensional feature values ​​of the reference vector. This embodiment uniformly stipulates a steady-state determination distance threshold of 0.25; d ≤ 0.25 is determined as a steady-state interval, and d > 0.25 is determined as a non-steady-state interval. The steady-state determination distance threshold is determined by statistically analyzing the Euclidean distance between the feature vectors of normal fluctuation segments in historical drilling data and the corresponding reference vectors, taking the 95th percentile value of the normal fluctuation distance distribution. This is for illustrative purposes only and is not intended to be limiting. During the period from 40 minutes and 10 seconds to 46 minutes and 10 seconds, the first deviations of six consecutive data segments are 0.08, 0.12, 0.10, 0.14, 0.09, and 0.13, respectively, and the second deviations are 0.06, 0.09, 0.08, 0.11, 0.07, and 0.10, respectively. All of these deviations are less than the steady-state determination distance of 0.25. The data from these six segments are then included in the dynamic benchmark sample set. The dynamic benchmark sample set uses a first-in, first-out (FIFO) storage rule, with a maximum storage capacity of... Ten data segments are used. If the data exceeds the capacity, the earliest stored data is automatically removed. The arithmetic mean of the sample set is used as the statistical center, and the dynamic benchmark is updated to [2.32, 0.78, 0.63]. In the data segment at 72 minutes and 10 seconds, the first deviation is 0.38, exceeding the steady-state judgment distance of 0.25, entering the non-steady-state interval. The dynamic benchmark update permission is immediately frozen, and sample set expansion and benchmark updates are suspended. In this embodiment, if the number of consecutive non-steady-state segments exceeds 10 during the freeze period, the sample set is cleared, pending re-establishment. After restoring steady state, accumulation resumes from the empty set. During the freeze period, the deviations of the seven consecutive segments from 73 minutes and 10 seconds to 79 minutes and 10 seconds all fall below 0.25. In this embodiment, the unfreezing condition is that there are no fewer than five consecutive steady-state segments, which means that statistical stability has been restored and the statistical stability required for benchmark updating is met. The freeze is then lifted, and the data of the above seven segments are included in the sample set. The statistical center is recalculated, and the dynamic benchmark is updated to [2.5, 0.9, 0.65]. This is only an example and is not a limitation.

[0023] Specifically, step S30 includes: S31. For each data segment, extract the characteristics of rod rotation smoothness, axial feed smoothness, and rod vibration frequency domain, respectively, as the feature values ​​of each dimension of the current data segment. S32. Calculate the deviation of each dimension feature value of the current data segment from the corresponding dimension feature value in the dynamic benchmark, and determine whether the deviation of each dimension is in the non-steady-state interval and continues to accumulate in the same direction. S33. When only one dimension of the deviation is in the unsteady range and drifts in one direction, while the deviations of other dimensions are still in the steady range, the stratigraphic trend component that changes monotonically with the footage depth is extracted from the current data segment. The extracted data is used to correct the characteristic value of the current data segment, and the deviation of each dimension is recalculated with the corrected characteristic value. Then, it is determined again whether the deviation of each dimension is in the unsteady range and continues to accumulate in the same direction. S34. When the deviations in all dimensions are in the unsteady range and all drift in one direction, calculate the first real-time correlation coefficient between the rotational smoothness feature and the axial feed smoothness feature in the current data segment, the second real-time correlation coefficient between the rotational smoothness feature and the vibration frequency domain feature of the rod, and the third real-time correlation coefficient between the axial feed smoothness feature and the vibration frequency domain feature of the rod. S35. Compare the first real-time correlation coefficient, the second real-time correlation coefficient, and the third real-time correlation coefficient with the corresponding correlation coefficients in the initial correlation coefficients, calculate the deviation of each correlation coefficient, and use the statistical mean of the deviation of each correlation coefficient as the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient.

[0024] Specifically, in this embodiment, the threshold for determining a single-dimensional unsteady state is determined by statistically analyzing the characteristic deviations of each dimension of the drill pipe under normal operating conditions in historical drilling data. It is taken as 30% of the corresponding dimension value of the dynamic benchmark. This proportion represents the relative deviation corresponding to three times the standard deviation of the normal deviation fluctuation range. If the absolute value of the deviation exceeds this threshold, the dimension is determined to have entered an unsteady drift state. This is only an example and is not a limitation. In this embodiment, unidirectional drift refers to the same sign of the deviation in the same dimension of adjacent data segments and an increasing trend in absolute value. Continuous accumulation refers to the continuous increase in the cumulative amount of unidirectional drift within at least three consecutive segments. For the data segment at 85 minutes and 10 seconds, the rod rotation stability feature is extracted. The value is 3.8, the axial feed smoothness characteristic value is 1.4, and the vibration frequency domain characteristic value is 0.51. The deviations between the current segment and the corresponding dimension characteristic values ​​in the dynamic reference are calculated: rotational smoothness deviation is |3.8-2.5|=1.3, feed smoothness deviation is |1.4-0.9|=0.5, and vibration frequency domain characteristic deviation is |0.51-0.65|=0.14. The unsteady-state judgment threshold is taken as 30% of the values ​​of each dimension in the dynamic reference, i.e., rotational smoothness threshold 2.5×0.3=0.75, feed smoothness threshold 0.9×0.3=0.27, and vibration frequency domain threshold 0.65×0.3=0.20. The rotational smoothness deviation of 1.3 is greater than 0.75. The feed smoothness deviations were greater than 0.27, both falling within the unsteady-state range. In the three consecutive data segments from 88 minutes 10 seconds to 91 minutes 10 seconds, only the rotational smoothness dimension showed a continuous unidirectional positive drift, with deviations of 0.35, 0.42, and 0.48 respectively, while the feed smoothness and vibration frequency domain characteristic deviations remained within the steady-state range. In this embodiment, the least squares method was used to linearly fit the rotational smoothness deviation and the footage depth, stripping away the formation trend component. The fitted linear relationship was Δ = a × D + b, where Δ is the rotational smoothness deviation, D is the footage depth, a is the fitting slope, and b is the fitting intercept. In this embodiment, the fitting results were a = 0.02 and b = 0. 15. The rotational stability deviation increases linearly with the advance depth at a rate of 0.02 per meter. The average residual after stripping is 0.03, close to zero. Since the single-dimensional drift is linearly related to the advance depth and the residual approaches zero, this drift is caused by lithological gradation and is not an equipment anomaly; therefore, it is determined to be caused by lithological gradation. The original deviation value is replaced with the residual after stripping, i.e., the remaining part after removing a×D+b from the rotational stability deviation Δ of each segment is used as the corrected rotational stability deviation. The deviations in the three dimensions are recalculated, and it is determined again whether the deviations in each dimension are all in the unsteady-state range and continuously accumulate in the same direction. The objective function of the least squares method is minΣ(Δi-a×Di-b). 2Where Δi is the rotational smoothness deviation of the i-th segment, Di is the feed depth corresponding to the i-th segment, a is the slope, b is the intercept, and Σ represents the summation over all segments involved in the fitting. In the eight consecutive data segments from 96 minutes and 10 seconds to 103 minutes and 10 seconds, the rotational smoothness deviations are 1.3, 1.5, 1.8, 2.1, 2.4, 2.7, 3.0, and 3.3, respectively; the feed smoothness deviations are 0.5, 0.6, 0.8, 0.9, 1.1, 1.3, 1.5, and 1.7, respectively; and the vibration frequency domain characteristic deviations are 0.08, 0.09, 0.12, 0.11, 0.15, 0.14, 0.18, and 0.17, respectively. All three dimensions are in the unsteady-state region and exhibit unidirectional drift. The real-time correlation of the three dimensions within the current segment is calculated. Coefficients; For the data segment at 103 minutes and 10 seconds, the first real-time correlation coefficient r12' is 0.28, the second real-time correlation coefficient r13' is -0.08, and the third real-time correlation coefficient r23' is 0.05. Comparing the three real-time correlation coefficients with the initial correlation coefficients [0.73, -0.41, -0.28], the deviation of the first correlation coefficient is |0.28-0.73|=0.45, the deviation of the second correlation coefficient is |-0.08-(-0.41)|=0.33, and the deviation of the third correlation coefficient is |0.05-(-0.28)|=0.33. The statistical mean of the three deviations is (0.45+0.33+0.33) / 3=0.37, which is used as the degree of offset of the current segment. This is only an example and is not a limitation.

[0025] Specifically, step S40 includes: S41. Obtain the degree of offset calculated from continuous data segments with statistical stability for trend determination, and form a time series of offset. S42. Determine whether the time series of offset shows a continuous increasing trend. A continuous increasing trend means that within the corresponding continuous data segment, the offset accumulates and increases in a single direction, and the accumulated increase exceeds the upper limit of the steady-state interval of the offset. S43. When the offset time series shows a continuous increasing trend, extract the per-pass depth change series corresponding to the continuous increasing trend, and determine whether there is a linear dependency between the increasing trend of offset and the per-pass depth change. S44. When there is no linear dependence between the increasing trend of the offset degree and the change in the depth of each pass, the offset degree change pattern judgment is triggered, and the monitoring of the offset degree change rate is initiated. S45. When there is a linear dependence between the increasing trend of the offset degree and the change of the depth of each pass, the change mode decision is not triggered. The drift trend of each dimension feature relative to the current dynamic benchmark continues to be monitored synchronously for subsequent data segments, and the offset degree of the real-time correlation coefficient between dimensions relative to the initial correlation coefficient is calculated.

[0026] Specifically, starting from 96 minutes and 10 seconds, the offset degrees of eight consecutive data segments are 0.15, 0.19, 0.23, 0.28, 0.31, 0.34, 0.37, and 0.40, forming an offset degree time series. The upper limit of the steady-state interval for offset degree is uniformly set to 0.12. In this embodiment, the upper limit of the steady-state interval is determined by statistically analyzing the offset degrees of consecutive segments under normal working conditions in historical drilling data and taking the 95th percentile value of the normal offset degree fluctuation range. This is only an example and is not a limitation. In the most recent eight segments, the offset degree has accumulated and increased in a single direction, with an accumulated increase of 0.40-0.15=0.25, exceeding the upper limit of the steady-state interval for offset degree of 0.12, and is judged to be a continuous increasing trend. The sequence of per-cycle depth changes corresponding to these eight segments is extracted. The depth of per-cycle increases uniformly from 312 meters to 336 meters, with each segment corresponding to an increase in depth of approximately 3 meters. The relationship between the increase in offset degree and the change in depth of per-cycle depth is determined. The linear dependence between the amplitudes was calculated using the offset increment sequence [0.04, 0.04, 0.05, 0.03, 0.03, 0.03, 0.03] and the footage increment sequence [3, 3, 3, 3, 3, 3, 3]. The Pearson correlation coefficient between these two sequences was 0.12, which is lower than the linearity threshold of 0.7. The linear correlation threshold is 0.7. A correlation coefficient ≥ 0.7 indicates the existence of stratigraphic linear dependence, while < 0.7 indicates no stratigraphic linear correlation. The threshold was determined based on the statistical significance standard of the Pearson correlation coefficient. In 8 sample pairs, a correlation coefficient of 0.7 corresponds to a p-value less than 0.05, indicating a statistically significant linear relationship between the two variables. This is only an example and is not a limitation. A Pearson correlation coefficient of 0.12 less than 0.7 indicates no linear dependence between the increase in offset and the change in footage depth, triggering a change pattern judgment and initiating monitoring of the rate of change in offset. This is only an example and is not a limitation.

[0027] Specifically, step S50 includes: S51. Obtain the offset degree of continuous data segments after the trigger change mode decision, and form an offset degree time series; calculate the first difference of the offset degree between adjacent data segments in the offset degree time series, and use the first difference to characterize the rate of change of the offset degree, and form a rate of change series. S52. Monitor the rate of change sequence and determine whether there is a step jump in the rate of change from the previous steady state level to the next higher steady state level. A step jump means that the increase in the rate of change between adjacent data segments exceeds the upper limit of the steady state range of the rate of change, and the rate of change after the jump remains at the new steady state level. S53. When a step jump is detected, record the time point corresponding to the starting data segment of the jump as the step time point. S54. Extract the drift curves of the drift of each dimension feature value relative to the current dynamic benchmark over time, and identify the trend turning points where the drift direction or drift slope changes on the drift curves of each dimension. S55. Determine whether the time deviation between the step time point and the trend turning point of the drift curve in each dimension is within the synchronous judgment interval. When the trend turning point of all dimensions is synchronized with the step time point, it is determined that a drill pipe misalignment abnormality has occurred.

[0028] Specifically, the offset of seven consecutive data segments from 103 minutes and 10 seconds to 109 minutes and 10 seconds after the trigger change mode decision is obtained, with offsets of 0.40, 0.43, 0.47, 0.53, 0.61, 0.71, and 0.83 respectively, forming an offset time series; the first-order difference between adjacent segments is calculated, resulting in a change rate sequence of 0.03, 0.04, 0.06, 0.08, 0.10, and 0.12; the change rate sequence is monitored, with steady-state level referring to the state where the change rate sequence fluctuates slightly around a certain central value within consecutive segments, and the fluctuation amplitude does not exceed the upper limit of the steady-state interval of the change rate; higher-order steady-state level refers to the state where the change rate re-circles after the jump. The system is characterized by a small fluctuation around a central value significantly higher than the previous steady-state level. A step jump is identified when the rate of change changes from a previous steady-state level of 0.03-0.04 to a higher-order steady-state level of 0.08-0.12, with a jump amplitude of 0.08, exceeding the upper limit of the steady-state range of 0.05. Furthermore, if the rate of change remains above 0.08 for three consecutive segments after the jump, a step jump is determined. In this embodiment, the upper limit of the steady-state range is determined by statistically analyzing the fluctuation range of the rate of change of offset under normal operating conditions in historical drilling data, and taking the 95th percentile value of the normal rate of change fluctuation range. This is only an example and not a limitation. The starting data segment corresponding to the jump is recorded. The time point is 106 minutes and 10 seconds, designated as the step time point. Drift curves are extracted from the three dimensions of feature values ​​relative to the dynamic baseline for each segment from 96 minutes and 10 seconds to 109 minutes and 10 seconds, showing the drift over time. The slope of the rotational smoothness drift curve changes from 0.3 to 0.6 at 106 minutes and 10 seconds; the slope of the feed smoothness drift curve changes from 0.15 to 0.35 at the same time point; and the slope of the vibration frequency domain drift curve changes from 0.02 to 0.06 at the same time point. All three dimensions show a trend inflection point at 106 minutes and 10 seconds. In this embodiment, the trend inflection point is identified using a sliding window slope change detection method, with a window length of three segments. The slope of the linear fit within the window is calculated. When the slope change rate between two adjacent windows exceeds 50%, it is marked as a trend inflection point. In this embodiment, the synchronization judgment interval is ±15 seconds, determined according to the duration of the data segment and the sliding window step size. 25% of the duration of a single data segment of 60 seconds is taken as the maximum allowable time deviation to ensure that the trend inflection point and the step time point are captured in the same segment. The step time point is 106 minutes and 10 seconds. The trend inflection points of all dimensions are also at 106 minutes and 10 seconds, and the time deviation is 0. They are all within the synchronization judgment interval. The trend inflection points of all dimensions are synchronized with the step time point, and the drill pipe misalignment anomaly is judged to have occurred. This is only an example and is not a limitation.

[0029] This invention provides another technical solution: an artificial intelligence-based anomaly monitoring system for exploration operations. The system includes: a baseline construction module, a dynamic refresh module, a coupling offset module, and an anomaly judgment module. The benchmark construction module is used to calibrate the return boundary based on the identification results of drill pipe unloading events, extract drill pipe assembly behavior data in the initial period of the return, and generate multi-dimensional benchmark features and initial correlation coefficients between dimensions to characterize the overall dynamic characteristics of the drill pipe assembly. The self-consistency verification of the initial period data is used as the effective condition for the multi-dimensional benchmark features and initial correlation coefficients. The dynamic refresh module is used to set the behavioral characteristics of the first data segment in a cycle as the dynamic benchmark and start the dynamic benchmark refresh cycle. When the deviations of subsequent data segments from the multidimensional benchmark features and from the current dynamic benchmark are both in the steady-state range, the corresponding data segments are included and the dynamic benchmark is updated. When any deviation enters the non-steady-state range, the update permission of the dynamic benchmark is immediately frozen. The coupling offset module is used to synchronously monitor the drift trend of each dimension feature relative to the current dynamic benchmark for each data segment. If only a single dimension shows unidirectional drift, the stratigraphic trend stripping is performed and the feature value of the current data segment is corrected before the drift monitoring is re-executed. If all dimensions show unidirectional drift, the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient is calculated. The anomaly detection module is used to trigger a change mode detection when the degree of offset continues to increase and its increasing trend has no linear dependence on the change in the depth of each pass, and to monitor the rate of change of the degree of offset; when the rate of change changes by a step jump and the step time point is synchronized with the trend inflection point of the drift curve of each dimension, it is determined that a drill pipe misalignment anomaly has occurred.

[0030] Specifically, the dynamic refresh module includes a baseline update unit and a baseline freeze unit; The benchmark update unit is used to include the data of the continuous data segment into the dynamic benchmark sample set when the first deviation of the continuous data segment from the multidimensional benchmark features and the second deviation from the current dynamic benchmark are both in the steady state range, and to update the dynamic benchmark with the statistical center of the dynamic benchmark sample set. The benchmark freeze unit is used to immediately freeze the dynamic benchmark update permission when the first or second deviation of any data segment enters the non-steady-state interval, suspending the expansion of the dynamic benchmark sample set and the update of the dynamic benchmark; when the data segment that has been continuously in the steady-state interval regains the statistical stability required for benchmark update, the freeze is lifted, and the data segment that has been continuously in the steady-state interval is re-included into the dynamic benchmark sample set, and the dynamic benchmark is updated again with the statistical center of the dynamic benchmark sample set.

[0031] Specifically, the coupled offset module includes a drift monitoring unit and an offset quantization unit; The drift monitoring unit is used to calculate the deviation of each dimension feature value of each data segment from the corresponding dimension feature value in the dynamic benchmark. When only the deviation of a single dimension is in the non-steady-state range and drifts in one direction, while the deviations of other dimensions are still in the steady-state range, the stratigraphic trend component that changes monotonically with the footage depth is extracted from the current data segment. The extracted data is used to correct the feature value of the current data segment, and the deviation of each dimension is recalculated with the corrected feature value. Then, it is determined again whether the deviation of each dimension is in the non-steady-state range and continues to accumulate in the same direction. The offset quantization unit is used to calculate the real-time correlation coefficients between the features of each dimension in the current data segment when the deviations of all dimensions are in the non-steady-state range and all are unidirectional drifts. It then compares each real-time correlation coefficient with the corresponding correlation coefficient in the initial correlation coefficient and uses the statistical mean of the deviation of each correlation coefficient as the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient.

[0032] Specifically, the anomaly detection module includes a trend verification unit and a step detection unit; The trend verification unit is used to obtain the offset degree calculated from continuous data segments with statistical stability for trend determination, forming an offset degree time series; determine whether the offset degree time series shows a continuous increasing trend; when a continuous increasing trend is shown, determine whether there is a linear dependency between the increasing trend of the offset degree and the change in the depth of each pass; when there is no linear dependency, trigger the change mode judgment and start the change rate monitoring; when there is a linear dependency, do not trigger the change mode judgment, and continue to synchronously monitor the drift trend of each dimension feature relative to the current dynamic benchmark for subsequent data segments and calculate the offset degree of the real-time correlation coefficient between dimensions relative to the initial correlation coefficient; The step detection unit is used to obtain the degree of offset of continuous data segments after the trigger change mode decision, calculate the first-order difference of the offset between adjacent data segments to characterize the rate of change; determine whether there is a step jump in the rate of change from the previous steady-state level to the next higher steady-state level; when a step jump is detected, record the time point corresponding to the starting data segment of the jump as the step time point; extract the drift curve of the drift of each dimension feature value of each data segment relative to the current dynamic benchmark over time, and identify the trend inflection points on the drift curves of each dimension; when the trend inflection points of all dimensions are synchronized with the step time point, it is determined that a drill pipe misalignment anomaly has occurred.

[0033] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An artificial intelligence-based method for monitoring anomalies in exploration operations, characterized by: The method includes: S10. Based on the identification results of drill pipe unloading events, mark the return boundary, extract the drill pipe assembly behavior data in the initial period of the return, generate multi-dimensional benchmark features and initial correlation coefficients between dimensions to characterize the overall dynamic characteristics of the drill pipe assembly, and take the self-consistency verification of the initial period data as the effective condition for the multi-dimensional benchmark features and initial correlation coefficients. S20. Set the behavior feature of the first data segment in the cycle as the dynamic benchmark and start the dynamic benchmark refresh cycle. When the deviations of subsequent data segments from the multidimensional benchmark feature and from the current dynamic benchmark are both in the steady state range, the corresponding data segment data is included and the dynamic benchmark is updated. When any deviation enters the non-steady state range, the update permission of the dynamic benchmark is immediately frozen. S30. For each data segment, synchronously monitor the drift trend of each dimension feature relative to the current dynamic benchmark; if only a single dimension shows unidirectional drift, perform stratigraphic trend stripping and correct the feature value of the current data segment before re-executing drift monitoring; if all dimensions show unidirectional drift, calculate the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient. S40. When the degree of offset continues to increase, and the increasing trend of the degree of offset has no linear dependence on the change of the depth of advance per pass, the change pattern judgment of the degree of offset is triggered. S50. Monitor the rate of change of the offset. When the rate of change changes by a step jump and the step time point is synchronized with the trend inflection point of the drift curve of each dimension, it is determined that a drill pipe misalignment anomaly has occurred.

2. The method for monitoring anomalies in exploration operations based on artificial intelligence according to claim 1, characterized in that: Step S10 includes: S11. Obtain the identification results of drill pipe unloading events. Take the pure drilling interval without unloading events as a single cycle, take the time of the last unloading completion in the cycle as the cycle start boundary, and take the time of the first unloading start in the cycle as the cycle end boundary. S12. From the first steady-state drilling period after the initial boundary of the cycle, extract continuous data with statistical stability as the initial period data. The steady-state drilling period refers to the continuous period in which the fluctuation range of drill pipe rotation speed and drilling pressure are both within the steady-state determination range. S13. Extract the rod rotation smoothness characteristics, axial feed smoothness characteristics, and rod vibration frequency domain characteristics of the drill rod assembly from the initial time period data to construct multi-dimensional benchmark characteristics; S14. Calculate the correlation coefficient between any two dimensions in the multidimensional benchmark feature, and use it as the initial correlation coefficient between dimensions; S15. Divide the initial time period data into several verification data segments according to time, calculate the multidimensional feature value in each verification data segment, and determine that the multidimensional benchmark feature and the initial correlation coefficient are effective when the deviation between the multidimensional feature values ​​of each verification data segment are within the same stable distribution range. If the verification fails, slide the data capture window backward, re-extract the initial time period data and complete the feature construction and consistency verification until the verification is successful.

3. The method for monitoring anomalies in exploration operations based on artificial intelligence according to claim 1, characterized in that: Step S20 includes: S21. After the initial boundary of the cycle, the multidimensional feature values ​​and interdimensional correlation coefficients extracted in the first data segment are set as dynamic benchmarks. The first data segment is a continuous data segment of the same length as the initial time period data, which is truncated from the end time of the initial time period data. S22. Start the dynamic benchmark refresh loop, take subsequent data segments in sequence, and calculate the first deviation between the multidimensional feature value of the current data segment and the multidimensional benchmark feature, and the second deviation between the current data segment and the dynamic benchmark. S23. When both the first deviation and the second deviation of the continuous data segment are in the steady state range, the data of the continuous data segment is included in the dynamic benchmark sample set, and the dynamic benchmark is updated with the statistical center of the dynamic benchmark sample set. S24. When the first deviation or the second deviation of any data segment enters the non-steady-state region, immediately freeze the update permission of the dynamic benchmark, and suspend the expansion of the dynamic benchmark sample set and the update of the dynamic benchmark. S25. Continuously monitor the first deviation and the second deviation of subsequent data segments. When a data segment that has been continuously in the steady-state range regains the statistical stability required for benchmark updates, unfreeze it and re-include the data segment that has been continuously in the steady-state range into the dynamic benchmark sample set. Update the dynamic benchmark again with the statistical center of the dynamic benchmark sample set.

4. The method for monitoring anomalies in exploration operations based on artificial intelligence according to claim 1, characterized in that: Step S30 includes: S31. For each data segment, extract the characteristics of rod rotation smoothness, axial feed smoothness, and rod vibration frequency domain, respectively, as the feature values ​​of each dimension of the current data segment. S32. Calculate the deviation between the feature values ​​of each dimension of the current data segment and the corresponding feature values ​​of the dynamic benchmark, and determine whether the deviation of each dimension is in the non-steady-state interval and continues to accumulate in the same direction. S33. When only one dimension of the deviation is in the unsteady range and drifts in one direction, while the deviations of other dimensions are still in the steady range, the stratigraphic trend component that changes monotonically with the footage depth is extracted from the current data segment. The extracted data is used to correct the characteristic value of the current data segment, and the deviation of each dimension is recalculated with the corrected characteristic value. Then, it is determined again whether the deviation of each dimension is in the unsteady range and continues to accumulate in the same direction. S34. When the deviations in all dimensions are in the unsteady range and all drift in one direction, calculate the first real-time correlation coefficient between the rotational smoothness feature and the axial feed smoothness feature in the current data segment, the second real-time correlation coefficient between the rotational smoothness feature and the vibration frequency domain feature of the rod, and the third real-time correlation coefficient between the axial feed smoothness feature and the vibration frequency domain feature of the rod. S35. The first real-time correlation coefficient, the second real-time correlation coefficient, and the third real-time correlation coefficient are compared with the corresponding correlation coefficients in the initial correlation coefficients, and the deviation of each correlation coefficient is calculated. The statistical mean of the deviation of each correlation coefficient is used as the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient.

5. The method for monitoring anomalies in exploration operations based on artificial intelligence according to claim 1, characterized in that: Step S40 includes: S41. Obtain the offset degree calculated from a continuous data segment with statistical stability for trend determination, and form an offset degree time series; S42. Determine whether the offset time series shows a continuous increasing trend. The continuous increasing trend refers to the offset accumulating in a single direction within the corresponding continuous data segment, and the accumulated increase exceeds the upper limit of the steady-state interval of the offset. S43. When the offset time series shows a continuous increasing trend, extract the per-pass depth change series corresponding to the continuous increasing trend, and determine whether there is a linear dependency between the increasing trend of the offset and the per-pass depth change. S44. When there is no linear dependence between the increasing trend of the offset degree and the change in the depth of each pass, the offset degree change pattern judgment is triggered, and the monitoring of the rate of change of the offset degree is initiated. S45. When there is a linear dependence between the increasing trend of the offset degree and the change of the depth of each pass, the change mode decision is not triggered. The drift trend of each dimension feature relative to the current dynamic benchmark continues to be monitored synchronously for subsequent data segments, and the offset degree of the real-time correlation coefficient between dimensions relative to the initial correlation coefficient is calculated.

6. The method for monitoring anomalies in exploration operations based on artificial intelligence according to claim 1, characterized in that: Step S50 includes: S51. Obtain the offset degree of consecutive data segments after the trigger change mode decision, and form an offset degree time series; calculate the first difference of the offset degree between adjacent data segments in the offset degree time series, and use the first difference to characterize the rate of change of the offset degree, and form a rate of change sequence. S52. Monitor the change rate sequence and determine whether there is a step jump in the change rate from the previous steady state level to the next higher steady state level. The step jump refers to the increase in the change rate between adjacent data segments exceeding the upper limit of the steady state range of the change rate, and the change rate after the jump is maintained at the new steady state level. S53. When the step jump is detected, record the time point corresponding to the starting data segment of the jump as the step time point. S54. Extract the drift curves of the drift of each dimension feature value relative to the current dynamic benchmark over time, and identify the trend turning points where the drift direction or drift slope changes on the drift curves of each dimension. S55. Determine whether the time deviation between the step time point and the trend turning point of each dimension drift curve is within the synchronization determination interval. When the trend turning points of all dimensions are synchronized with the step time point, determine that a drill pipe misalignment abnormality has occurred.

7. An artificial intelligence-based anomaly monitoring system for exploration operations, characterized in that: The system includes: a baseline construction module, a dynamic refresh module, a coupling offset module, and an anomaly detection module; The benchmark construction module is used to calibrate the return boundary based on the identification results of drill pipe unloading events, extract drill pipe assembly behavior data in the initial period of the return, generate multi-dimensional benchmark features and initial correlation coefficients between dimensions that characterize the overall dynamic characteristics of the drill pipe assembly, and use the self-consistency verification of the initial period data as the effective condition for the multi-dimensional benchmark features and initial correlation coefficients. The dynamic refresh module is used to set the behavioral characteristics of the first data segment in a cycle as the dynamic benchmark and start the dynamic benchmark refresh cycle; when the deviations of subsequent data segments from the multidimensional benchmark features and from the current dynamic benchmark are both in the steady state range, the corresponding data segments are included and the dynamic benchmark is updated; when any deviation enters the non-steady state range, the update permission of the dynamic benchmark is immediately frozen. The coupling offset module is used to synchronously monitor the drift trend of each dimension feature relative to the current dynamic benchmark for each data segment; if only a single dimension shows unidirectional drift, the stratigraphic trend stripping is performed and the feature value of the current data segment is corrected before the drift monitoring is re-executed; if all dimensions show unidirectional drift, the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient is calculated. The anomaly detection module is used to trigger a change mode detection when the degree of offset continues to increase and its increasing trend has no linear dependence on the change in the depth of each pass, and to monitor the rate of change of the degree of offset; when the rate of change undergoes a step jump and the step time point is synchronized with the trend inflection point of the drift curve of each dimension, it is determined that a drill pipe misalignment anomaly has occurred.

8. The artificial intelligence-based anomaly monitoring system for exploration operations according to claim 7, characterized in that: The dynamic refresh module includes a baseline update unit and a baseline freeze unit; The benchmark update unit is used to include the data of the continuous data segment into the dynamic benchmark sample set when both the first deviation of the continuous data segment from the multidimensional benchmark feature and the second deviation from the current dynamic benchmark are in the steady state range, and to update the dynamic benchmark with the statistical center of the dynamic benchmark sample set. The benchmark freezing unit is used to immediately freeze the update permission of the dynamic benchmark when the first deviation or the second deviation of any data segment enters the non-steady-state interval, and to suspend the expansion of the dynamic benchmark sample set and the update of the dynamic benchmark; when the data segment that has been continuously in the steady-state interval regains the statistical stability required for benchmark update, the freeze is lifted, and the data segment that has been continuously in the steady-state interval is re-included into the dynamic benchmark sample set, and the dynamic benchmark is updated again with the statistical center of the dynamic benchmark sample set.

9. The artificial intelligence-based anomaly monitoring system for exploration operations according to claim 7, characterized in that: The coupling offset module includes a drift monitoring unit and an offset quantization unit; The drift monitoring unit is used to calculate the deviation of each dimension feature value of each data segment from the corresponding dimension feature value in the dynamic benchmark. When only the deviation of a single dimension is in the unsteady range and drifts in one direction while the deviations of other dimensions are still in the steady range, the stratigraphic trend component that changes monotonically with the footage depth is extracted from the current data segment. The extracted data is used to correct the feature value of the current data segment, and the deviation of each dimension is recalculated with the corrected feature value. The unit then determines again whether the deviation of each dimension is in the unsteady range and continues to accumulate in the same direction. The offset quantization unit is used to calculate the real-time correlation coefficient between the features of each dimension in the current data segment when the deviations of all dimensions are in the non-steady-state range and all are unidirectional drifts, and compare each real-time correlation coefficient with the corresponding correlation coefficient in the initial correlation coefficient, and use the statistical mean of the deviation of each correlation coefficient as the degree of deviation of the real-time correlation coefficient between the dimensions of the current data segment relative to the initial correlation coefficient.

10. The artificial intelligence-based anomaly monitoring system for exploration operations according to claim 7, characterized in that: The anomaly detection module includes a trend verification unit and a step detection unit; The trend verification unit is used to obtain the offset degree calculated from continuous data segments with statistical stability of trend determination, and form an offset degree time series; determine whether the offset degree time series shows a continuous increasing trend; when it shows a continuous increasing trend, determine whether there is a linear dependence between the increasing trend of the offset degree and the change in the depth of each advance. When there is no linear dependency, a change mode decision is triggered and change rate monitoring is started; when there is a linear dependency, a change mode decision is not triggered, and the drift trend of each dimension feature relative to the current dynamic benchmark is continuously monitored synchronously for subsequent data segments, and the deviation of the real-time correlation coefficient between dimensions relative to the initial correlation coefficient is calculated. The step determination unit is used to acquire the offset degree of continuous data segments after the trigger change mode judgment, calculate the first-order difference of the offset degree between adjacent data segments to characterize the change rate; determine whether there is a step jump in the change rate from the previous steady-state level to the next higher steady-state level; when a step jump is detected, record the time point corresponding to the starting data segment of the jump as the step time point; extract the drift curve of the drift amount of each dimension feature value of each data segment relative to the current dynamic benchmark over time, and identify the trend turning points on the drift curves of each dimension; when the trend turning points of all dimensions are synchronized with the step time point, it is determined that a drill pipe misalignment anomaly has occurred.

Images