A radar monitoring and evaluation method for safety situation of a shale gas well site

Abstract

The application provides a radar monitoring and evaluation method for a shale gas well site safety situation, comprising the following steps: collecting original vibration amplitude data, initial clutter power spectrum information and target Doppler shift signals, and obtaining filtered vibration amplitude data, purified clutter spectrum data and refined Doppler shift values after filtering; performing clustering analysis on the current vibration intensity level of the energy distribution characteristics of different frequency bands in the overlapping interval range, evaluating the interference level caused by the target detection of the clutter power spectrum broadening amount, and obtaining the interference intensity grading result; performing boundary feature extraction and trend fitting processing on the interference intensity grading result and the frequency spectrum overlapping boundary by using a support vector machine, and identifying a dynamic threshold adjustment amplitude to determine an adjusted detection threshold value.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a radar monitoring and assessment method for the safety status of shale gas well sites. Background Technology

[0002] Safety monitoring of shale gas well sites is a crucial research area, directly impacting the safety and stability of energy extraction. In complex well site environments, ensuring perimeter security and timely detection of potential intrusions are critical not only for on-site operational safety but also for equipment operation and resource protection. Real-time safety assessment is particularly urgent under high-intensity conditions such as fracturing operations. However, current safety monitoring methods often reveal shortcomings when dealing with complex environments. Many solutions fail to adequately consider the dynamic interference from well site operations, such as the impact of equipment vibration on monitoring signals, making it difficult for the system to accurately distinguish between real threats and environmental noise when faced with various interferences. This deficiency stems not merely from technical limitations but from a lack of in-depth analysis of the complex relationship between environmental changes and target behavior, resulting in significantly reduced monitoring effectiveness in specific scenarios. A deeper technical challenge lies in finding a balance between background interference caused by equipment vibration and the weak signals of intruding targets. Vibration interference significantly alters the characteristics of surface signals, causing a shift in signal frequency distribution, which often closely resembles the signal characteristics of some low-speed moving targets. This is because the periodic reciprocating motion of high-power pump sets during shale gas fracturing operations generates predominantly low-frequency vibration energy on the surface, with the dominant frequency typically concentrated in the 0.5 Hz to 5 Hz range. The Doppler frequency shift generated by the slow, intermittent contact of an intruding target with the ground also falls within this frequency range, creating a natural overlap between the two types of signals in the frequency domain. Due to the failure to effectively address the overlap between vibration interference and target signals, the system often faces a dilemma in actual operation: setting high detection standards to reduce false alarms caused by interference may overlook genuine low-speed intrusion behavior; conversely, lowering the standards to capture weak signals may lead to frequent false alarms due to interference. Therefore, in practical well site perimeter security monitoring, accurately distinguishing the subtle differences between low-speed moving intruding targets and background interference under conditions of strong vibration generated by equipment operation has become a critical problem that urgently needs to be solved. Taking a typical scenario during the fracturing construction phase of a shale gas well site as an example, when multiple fracturing pump sets are simultaneously performing high-pressure injection operations, the surface vibration intensity at the well site can reach several times that of normal conditions. Vibration energy concentrates and expands towards the low-frequency region. If an intruding target slowly approaches the well site perimeter in an intermittent manner, the signal characteristics generated by its movement almost overlap with the fracturing vibration interference, making it difficult for the system to distinguish between a real threat and environmental noise. This contradiction directly affects the reliability of safety monitoring. As the above analysis shows, the problem of distinguishing between vibration interference and target signals permeates the entire process of well site safety monitoring. It is urgent to find an effective solution to this core contradiction, thereby improving the system's adaptability in complex environments. Summary of the Invention

[0003] This invention provides a radar monitoring and assessment method for the safety status of shale gas well sites, mainly including: The original vibration amplitude data, initial clutter power spectrum information, and target Doppler frequency shift signal are collected, and after filtering, the filtered vibration amplitude data, purified clutter spectrum data, and refined Doppler frequency shift value are obtained. The current vibration intensity level is extracted from the filtered vibration amplitude data. At the same time, the power spectrum broadening of clutter is identified from the cleaned clutter spectrum data. The overlap range is determined by analyzing the spectrum overlap boundary between the low-speed target and the vibration clutter in combination with the refined Doppler frequency shift value. Cluster analysis is performed on the energy distribution characteristics of different frequency bands within the overlapping interval to determine the current vibration intensity level, assess the interference level caused by the clutter power spectrum broadening to target detection, and obtain the interference intensity classification results. Support vector machine is used to extract boundary features and fit trends between the interference intensity classification results and the spectrum overlap boundary, and the dynamic threshold adjustment range is identified to determine the adjusted detection threshold value. Feature extraction is performed on the positional distribution of refined Doppler frequency shift values ​​within the spectral overlap boundary. The embedding depth is obtained by assessing the depth to which the target signal is embedded in the clutter region. Preliminary target type determination results are obtained by identifying intrusive targets that move slowly and intermittently on the ground. Process the preliminary target type determination results of multiple consecutive frames, distinguish the mechanical vibration and intermittent ground contact movement characteristics caused by drilling equipment or fracturing truck operations based on the real-time changes of the current vibration intensity level, and track the motion posture of low-speed targets to determine the stable intrusion identification results. By combining historical cumulative data on clutter power spectrum broadening with stable intrusion identification results, the current false alarm rate and false negative rate are assessed. The dynamic threshold adjustment range is iteratively optimized to obtain the final optimized threshold value, thereby achieving effective identification of low-speed, intermittent, ground-contacting intrusion targets in shale gas well sites.

[0004] Furthermore, raw vibration amplitude data, initial clutter power spectrum information, and target Doppler frequency shift signal are collected, filtered to obtain filtered vibration amplitude data, cleaned clutter spectrum data, and refined Doppler frequency shift values, including: Millimeter-wave radar probes are deployed at preset intervals around the perimeter of the shale gas well site. Each probe is equipped with a vibration sensing unit to extract initial clutter power spectrum information from the radar echo and simultaneously capture the original Doppler frequency shift signal generated by target movement to obtain original vibration amplitude data. The original vibration amplitude data is then subjected to low-pass filtering to obtain filtered vibration amplitude data. The initial clutter power spectrum information is then filtered to obtain purified clutter spectrum data. Finally, the original Doppler frequency shift signal is subjected to frequency domain narrowband filtering to extract the effective frequency shift component and obtain the refined Doppler frequency shift value.

[0005] Furthermore, the step of extracting the current vibration intensity level based on the filtered vibration amplitude data, identifying the clutter power spectral broadening from the cleaned clutter spectral data, and determining the overlap range by analyzing the spectral overlap boundary between the low-speed target and the vibration clutter using refined Doppler frequency shift values ​​includes: The energy integral value of the vibration signal within the time window is calculated based on the filtered vibration amplitude data and compared with the multi-level vibration threshold to determine the current vibration intensity level. The power spectral density function is extracted from the cleaned clutter spectrum data, and the power spectral frequency width is calculated to obtain the clutter power spectral broadening. A clustering algorithm is used to cluster the overlapping parts by spectral energy density. The number of cluster centers is adjusted according to the current vibration intensity level to identify the overlapping start and end frequencies of the vibration clutter and the target frequency shift, and to determine the overlapping interval range.

[0006] Furthermore, the method also includes: obtaining the periodic energy pulse waveform of the pump group's reciprocating motion from the purified clutter spectrum data, acquiring the intermittent frequency shift waveform of the discontinuous ground contact moving target from the refined Doppler frequency shift value, analyzing the dominant position of the pump group's vibration energy pulse in the low frequency band, evaluating the distribution position of the discontinuous ground contact motion frequency shift waveform in the same low frequency band, determining the overlap boundary range of the two types of waveforms in the frequency band, and identifying the strength contrast between the pump group's energy pulse and the discontinuous ground contact signal within the overlap area.

[0007] Furthermore, identifying the strength contrast between the pump energy pulses and the intermittent ground contact signal within the overlapping area includes: obtaining the vibration energy envelope curve through short-time Fourier transform to obtain the periodic energy pulse waveform of the pump's reciprocating motion; detecting the frequency shift abrupt change generated at the moment the target touches the ground to construct the intermittent frequency shift waveform of the intermittently touching moving target; and comparing the power spectral density values ​​of the periodic energy pulse waveform and the intermittent frequency shift waveform in the same low-frequency band to determine the overlapping boundary and the strength contrast state.

[0008] Furthermore, cluster analysis is performed on the energy distribution characteristics of different frequency bands within the overlapping interval to assess the current vibration intensity level, evaluate the interference level caused by clutter power spectrum broadening to target detection, and obtain interference intensity classification results, including: The spectral data within the overlapping interval is segmented according to the frequency step size. Energy density, peak frequency, and bandwidth parameters are extracted from each frequency band to form a feature vector. A hierarchical clustering algorithm is used to group these feature vectors, and the within-group variance of different clustering levels is calculated to determine the optimal number of clusters, obtaining energy distribution clustering results for different frequency bands. Based on the energy values ​​of each cluster center in the energy distribution clustering results, a correlation coefficient is calculated between the energy value and the current vibration intensity level, establishing a mapping relationship between vibration intensity and clustering features. The power decrease rate is calculated from the clutter power spectrum broadening as a broadening feature, and the ratio of the frequency range covered by the broadening feature to the target frequency band is used as a coverage metric. The coverage metric is compared with an interference assessment threshold to determine weak, moderate, or strong interference. The judgment boundary for each interference level is corrected based on the correlation coefficient in the mapping relationship, resulting in a corrected interference level. Cross-statistics are performed between the corrected interference level and the energy distribution clustering results to construct a correlation matrix between interference intensity and detection performance. Principal component analysis is used to identify the main interference factors, outputting the interference intensity classification result.

[0009] Furthermore, a support vector machine is used to extract boundary features and fit trends between the interference intensity classification results and the spectral overlap boundary, identifying the dynamic threshold adjustment range and determining the adjusted detection threshold value, including: The interference intensity classification results are converted into numerical feature vectors. Boundary start frequency, termination frequency, and boundary width parameters are extracted from the spectral overlap boundary to construct a training sample set. A support vector machine is used for classification training to obtain a decision function. A distance value sequence is calculated based on the decision function, and trend fitting is performed on the distance value sequence to obtain the slope parameter of the boundary change trend. A baseline value for threshold adjustment is calculated based on the slope parameter and the current interference intensity classification results. The baseline value is multiplied by a unit adjustment amount to obtain the dynamic threshold adjustment amplitude. The original detection threshold value is read from the radar detection parameters. The dynamic threshold adjustment amplitude is added to the original detection threshold value. The adjusted value is scaled according to the decision probability of the decision function, and the adjusted detection threshold value is output.

[0010] Furthermore, feature extraction is performed on the positional distribution of the refined Doppler frequency shift values ​​within the spectral overlap boundary. The embedding depth is obtained by assessing the depth to which the target signal is embedded in the clutter region. Preliminary target type determination results are obtained by identifying intrusive targets that move slowly and intermittently on the ground, including: The distribution of the refined Doppler frequency shift value within the spectral overlap boundary is statistically analyzed, and the frequency of occurrence, duration, and interval period of the frequency shift peak are extracted as location distribution features to obtain the spatial distribution parameters of the target signal. The ratio of the target signal power to the clutter power is calculated based on the spatial distribution parameters, and the signal-to-noise ratio is obtained by logarithmic operation of the ratio to determine the embedding depth. The target occurrence pattern is determined based on the proportion of duration to total observation time. The embedding depth is compared with the adjusted detection threshold to determine whether the target signal is separable. The intermittent ground contact motion characteristics are identified based on the interval period, and the target is identified as a low-speed, intermittent ground contact intrusion target. The preliminary target type determination result is output.

[0011] Furthermore, the system processes the preliminary target type determination results across multiple consecutive frames, distinguishes between mechanical vibrations and intermittent ground contact movement characteristics caused by drilling equipment or fracturing truck operations based on real-time changes in the current vibration intensity level, and tracks the motion posture of low-speed targets to determine stable intrusion identification results, including: The system acquires the preliminary target type determination results for multiple consecutive frames, counts the number of consecutive frames with the same determination result, and extracts the target's position coordinate sequence in each frame as a temporal feature. Based on the temporal feature, the system calculates the target's moving speed, monitors the real-time changes in the current vibration intensity level, and determines whether it is a mechanical vibration interference or a real intrusion target. The system performs trajectory smoothing on the position coordinate sequence of the real intrusion target, confirms the intrusion behavior, and outputs a stable intrusion identification result.

[0012] Furthermore, by combining historical cumulative data on clutter power spectrum broadening with stable intrusion identification results, the current false alarm rate and false negative rate are assessed. The dynamic threshold adjustment amplitude is iteratively optimized to obtain the final optimized threshold value, enabling effective identification of low-velocity, intermittent, ground-contact moving intrusion targets in shale gas well sites, including: The mean and variance of clutter power spectrum broadening are extracted from historical cumulative data. The false alarm rate is obtained by dividing the number of false alarm events by the total number of detections. The false detection rate is obtained by dividing the number of missed detections by the number of real targets from stable intrusion identification results. A performance evaluation index is constructed. Based on the performance evaluation index, the gradient descent method is used to calculate the balance point that minimizes the sum of the false alarm rate and the false detection rate to update the threshold value and obtain the optimized threshold parameter. The optimized threshold parameter is used to detect radar signals. When the signal power exceeds the threshold and the motion characteristics conform to the low-speed intermittent ground contact mode, it is marked as a suspected target. The stability of the suspected target in the continuous observation period is verified, and the final identification result of the low-speed intermittent ground contact moving intrusion target in the shale gas well field is output.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a radar monitoring and assessment method for the safety status of shale gas well sites, aiming to solve the detection interference problem caused by the overlap of vibration clutter and target signal spectra in complex environments. This invention deploys millimeter-wave radar and vibration sensing units to collect surface vibration data and Doppler frequency shift signals. After filtering, the vibration intensity level, clutter power spectrum broadening, and spectral overlap boundaries are extracted. Combined with frequency band overlap analysis of pump vibration and intermittent ground contact signals, the interference intensity is assessed and the detection threshold is dynamically adjusted. This invention utilizes support vector machines for boundary feature extraction and trend fitting to determine the embedding depth and separability of target signals. Finally, through multi-frame data processing and vibration intensity changes, mechanical vibration and intrusion behavior are distinguished, and the threshold is iteratively optimized to reduce false alarm and false detection rates. This invention achieves efficient identification of low-speed, intermittent ground contact moving targets, significantly improving the accuracy and reliability of shale gas well site security monitoring. Attached Figure Description

[0014] Figure 1 This is a flowchart of a radar monitoring and assessment method for the safety status of shale gas well sites according to the present invention.

[0015] Figure 2 This is a schematic diagram of a radar monitoring and assessment method for the safety status of shale gas well sites according to the present invention.

[0016] Figure 3 This is another schematic diagram of a radar monitoring and assessment method for the safety status of shale gas well sites according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0018] like Figures 1-3 This embodiment of a radar monitoring and assessment method for the safety status of shale gas well sites may specifically include: S101. Collect the original vibration amplitude data, initial clutter power spectrum information, and target Doppler frequency shift signal. After filtering, obtain the filtered vibration amplitude data, purified clutter spectrum data, and refined Doppler frequency shift value.

[0019] Millimeter-wave radar probes are deployed at preset intervals around the perimeter of the shale gas well site. Each probe is equipped with a vibration sensing unit. The radar probes emit electromagnetic waves to cover the perimeter area, and the vibration sensing units record the surface vibration acceleration values ​​at a preset sampling rate. Initial clutter power spectrum information, including the operating noise of the fracturing equipment, is extracted from the radar echo. Simultaneously, the original Doppler frequency shift signal generated by target movement is captured to obtain the original vibration amplitude data. The original vibration amplitude data is processed by Butterworth low-pass filtering to remove high-frequency noise interference, resulting in filtered vibration amplitude data. For the initial clutter power spectrum information, a Kalman filter is used to eliminate random noise components and retain the dominant clutter frequency component, obtaining purified clutter spectrum data. The original Doppler frequency shift signal is then processed by frequency domain narrowband filtering to extract the effective frequency shift component in the low-frequency band, resulting in the narrowband filtered Doppler frequency shift value. The periodic components of the equipment vibration are separated from the purified clutter spectrum data by comparing the filtered vibration amplitude data with the purified clutter spectrum data. Non-periodic target motion characteristics are identified from the narrowband filtered Doppler frequency shift value. The refined Doppler frequency shift value is obtained by spectrum difference processing, wherein the spectrum difference processing is performed by subtracting the spectrum of the filtered vibration amplitude data from the purified clutter spectrum to calculate the difference and obtain the refined Doppler frequency shift value.

[0020] In one implementation, the millimeter-wave radar probe at the perimeter of the shale gas well site employs a frequency-modulated continuous wave (FM) system. It transmits a linear frequency-modulated signal and receives the reflected echo from the target to achieve continuous monitoring of the perimeter area. The vibration sensing unit uses a piezoelectric accelerometer with a sensitivity range covering 0.01 to 1 kHz, effectively capturing low-frequency vibration signals generated during fracturing equipment operation. When the fracturing pump unit performs high-pressure injection, the reciprocating motion of the plunger generates periodic vibrations on the ground surface. These vibrations are transmitted to the sensor through the foundation, forming background noise with fixed frequency characteristics. The order of the Butterworth low-pass filter needs to be determined based on the actual vibration characteristics at the fracturing site. In the initial stage of fracturing operations, the impact vibrations generated during pump start-up contain a large number of high-frequency components. These transient interferences can mask the true motion characteristics of the target. By setting an appropriate cutoff frequency, the filter can effectively suppress sudden interference caused by equipment start-up and shutdown while retaining the target's low-speed motion information. The Kalman filter is used to process the random noise component in the clutter power spectrum. Its state equation is established based on the operation mode of the fracturing equipment, and the observation equation reflects the actual clutter power received by the radar. The purified clutter spectrum data is obtained by recursive estimation.

[0021] During the spectral differential processing, the periodic components manifest as energy concentrations at specific frequency points, which are directly related to the rotational speed of the fracturing pump unit.

[0022] For example, when a pump unit operates at a speed of 300 revolutions per minute, its fundamental frequency is 5 Hz, which appears as a distinct spectral line on the spectrum. The Doppler frequency shift caused by the intermittent ground contact motion of an invading target is manifested as a continuous broadening of the spectrum, and its frequency distribution depends on the target's moving speed and ground contact frequency.

[0023] In one possible implementation, the process of refining the Doppler frequency shift values ​​involves further processing of the spectral difference results. By calculating the rate of change of the spectrum within adjacent time windows, non-periodic frequency shift components are identified, corresponding to the actual motion state of the target. When the target approaches the well site at a slow speed, the resulting Doppler frequency shift, although weak, exhibits a different variation pattern in the time domain than equipment vibration, thus enabling effective differentiation between the two types of signals.

[0024] S102. Extract the current vibration intensity level based on the filtered vibration amplitude data. At the same time, identify the clutter power spectrum broadening from the cleaned clutter spectrum data, and combine the refined Doppler frequency shift value to analyze the spectral overlap boundary between the low-speed target and the vibration clutter to determine the overlap range.

[0025] The energy integral value of the vibration signal within the time window is calculated based on the filtered vibration amplitude data. This energy integral value is compared with a preset multi-level vibration threshold to determine the current vibration intensity level, and the frequency distribution characteristics corresponding to the vibration intensity level are recorded. The power spectral density function is extracted from the purified clutter spectrum data, and the frequency width corresponding to the drop in power spectrum from the peak value to half the peak value in the frequency domain is calculated to obtain the clutter power spectral broadening. The upper and lower limits of the frequency shift generated by the target motion are identified based on the refined Doppler frequency shift value after filtering optimization. The frequency band covered by the clutter power spectral broadening is compared with the frequency band of the Doppler frequency shift. The K-means clustering algorithm is used to perform spectral energy density clustering on the overlapping parts in the frequency band comparison results. The number of cluster centers is adjusted according to the current vibration intensity level. The overlapping start and end frequencies of the vibration clutter and target frequency shift in the frequency domain are identified, the spectral overlap boundary is determined, and the overlap interval range is calculated.

[0026] In one implementation, the vibration energy integral value is calculated by integrating the filtered vibration amplitude data within a preset time window. The time window length is set to one second, and the squares of the vibration amplitudes are accumulated within this window to obtain the energy value reflecting the vibration intensity. Multi-level vibration thresholds are preset according to different stages of fracturing operations. In the initial stage of fracturing, the vibration energy integral value of a single pump unit is usually at a low level. When multiple pump units work simultaneously, the vibration energy exhibits a stepwise increase.

[0027] Specifically, the power spectral density function is obtained by performing a fast Fourier transform on the cleaned clutter spectrum data, and it reflects the energy distribution on different frequency components.

[0028]

[0029] PSD(f) represents the power spectral density function, f represents the frequency, and FFT{s(n)} represents the fast Fourier transform of the cleaned clutter spectrum data s(n), with the squared modulus yielding the energy distribution at different frequency components. The physical significance of clutter power spectral broadening lies in characterizing the range of influence of vibration interference on radar detection. When the vibration of fracturing equipment intensifies, the clutter energy originally concentrated at a specific frequency will diffuse to both sides, forming a spectral broadening phenomenon. The half-power bandwidth is measured starting from the peak of the power spectrum, searching for frequency points where the power drops to half the peak value on both sides; the frequency difference between the two points is the broadening. In this application, the K-means clustering algorithm clusters the spectral energy density value. The algorithm groups the spectral data according to the similarity of energy density, with each group representing a signal source. The vibration intensity level directly affects the number of cluster centers. High-intensity vibrations involve complex clutter components, requiring more cluster centers to distinguish different vibration sources. By analyzing the frequency distribution of each cluster center, the frequency components belonging to the fracturing equipment vibration and the frequency components belonging to the target motion are identified, and the boundary between the two is the spectral overlap boundary.

[0030] Preferably, the determination of the overlapping interval range also considers changes in the time dimension. At different stages of fracturing operations, the equipment's operating state changes, leading to corresponding changes in vibration characteristics and spectral distribution. By continuously monitoring the spectral overlap within multiple time windows, a dynamic change model of the overlapping interval is established, providing a basis for subsequent adjustment of the target detection threshold.

[0031] In one possible implementation, when the vibration intensity level jumps from a low level to a high level, the spectral overlap interval expands rapidly. At this time, the system records the expansion rate and the final stable range of the overlap interval. These parameters reflect the transition characteristics of the well site environment from a quiet state to a high-intensity operating state.

[0032] The periodic energy pulse waveform of the pump group's reciprocating motion was obtained from the purified clutter spectrum data. The intermittent frequency shift waveform of the intermittent ground contact moving target was collected from the refined Doppler frequency shift value. The dominant position of the pump group's vibration energy pulse in the low frequency band was analyzed. The distribution position of the intermittent ground contact motion frequency shift waveform in the same low frequency band was evaluated. The overlap boundary range of the two types of waveforms in the frequency band was determined. The strength contrast between the pump group's energy pulse and the intermittent ground contact signal in the overlap area was identified.

[0033] The periodic vibration components generated by the reciprocating motion of the pump plunger are extracted from the cleaned clutter spectrum data. The envelope curve of the vibration energy over time is obtained through short-time Fourier transform, and the periodic occurrence pattern of the envelope peak is identified to obtain the periodic energy pulse waveform of the pump reciprocating motion and its dominant frequency point in the low-frequency band. Based on the time-domain variation characteristics of the refined Doppler frequency shift value, the frequency shift abrupt change generated at the moment of target contact with the ground is detected. The frequency shift amplitude and interval of each contact action are recorded to construct the intermittent frequency shift waveform of the intermittently contacting moving target, and the distribution position of the intermittent frequency shift waveform near the dominant frequency point is evaluated. The power spectral density values ​​of the periodic energy pulse waveform and the intermittent frequency shift waveform in the same low-frequency band are compared. The overlap boundary is determined by calculating the frequency range of overlap between the two power spectra. The ratio of the pump energy pulse power to the intermittent contact signal power within the overlap area is used as the strength comparison state.

[0034] In one implementation, the short-time Fourier transform is achieved through time-frequency analysis of the cleaned clutter spectrum data. A Hanning window is selected as the window function, with a window length of 0.5 seconds and an overlap rate of 50%. Fourier transforms are performed on the data within each time window to obtain the spectral distribution for that period. The reciprocating motion of the pump plunger exhibits a regular energy concentration on the spectrum, with its period directly related to the pump rotation speed. When the pump operates at a constant speed, the energy envelope shows stable periodic fluctuations. The determination of the dominant frequency point depends on the peak detection of the power spectral density. In the low-frequency range, typically 0 to 10 Hz, local maxima of the power spectral density are identified; the frequencies corresponding to these maxima are the dominant frequencies of the pump vibration. In fracturing operations, different pump models will generate different dominant frequencies. The dominant frequencies of large fracturing pumps are usually concentrated between 2 and 5 Hz, while the dominant frequencies of auxiliary pumps are distributed between 5 and 8 Hz.

[0035] Specifically, the frequency shift characteristics of intermittently moving targets differ significantly from those of continuously moving targets. When an intruder approaches the well site by crawling or creeping, the ground vibrations caused by its alternating hand and foot contact with the ground will cause intermittent frequency shifts in the radar echo. At each contact moment, the relative velocity between the target and the ground changes abruptly, which manifests as an instantaneous frequency shift peak in the Doppler spectrum. During the contact interval, the frequency shift value returns to the baseline level, forming a frequency shift waveform with obvious intermittent characteristics. The period of this intermittent frequency shift waveform is typically 0.8 to 2 seconds, depending on the intruder's movement speed and contact rhythm.

[0036] Preferably, the power spectral density values ​​are compared using a decibel difference method for quantification. The power spectral density of the pump energy pulse is converted into decibel values, and the power spectral density of the intermittent ground contact signal is processed in the same way. The difference between the two reflects the relative relationship of signal strength. When the difference exceeds ten decibels, it indicates that pump vibration is dominant and the target signal is severely masked; when the difference is between three and ten decibels, the two types of signal strengths are comparable, and there is mutual interference; when the difference is less than three decibels, the target signal has the potential to be detected.

[0037] In one possible implementation, the determination of the overlapping boundary also considers the spectral leakage effect. Due to the non-ideal characteristics of actual signals, the energy of pump vibration will diffuse to adjacent frequencies, forming spectral leakage. This leakage causes the two originally separate signals to overlap in the frequency domain, and the width of the overlapping region is proportional to the vibration intensity.

[0038] S103. Perform cluster analysis on the energy distribution characteristics of different frequency bands within the overlapping interval to determine the current vibration intensity level, assess the interference level caused by the clutter power spectrum broadening to target detection, and obtain the interference intensity classification result.

[0039] The spectral data within the overlapping interval is segmented according to a preset frequency step size. The energy density, peak frequency, and bandwidth parameters of each frequency band are extracted to form a feature vector. A fully linked hierarchical clustering algorithm is used to group these feature vectors. By calculating the within-group variance W(k) of different clustering levels, the inflection point of the growth rate is determined. To obtain the optimal number of clusters, where W(k) is the within-group variance for cluster k, energy distribution clustering results for different frequency bands are obtained. Based on the energy values ​​of each cluster center in the energy distribution clustering results, the correlation coefficient between the energy value and the current vibration intensity level is calculated, establishing a mapping relationship between vibration intensity and clustering characteristics. The vibration intensity level is a preset value from 0 to 10. Simultaneously, the power decrease rate is calculated from the clutter power spectrum broadening (spectral width difference) calculated in the previous step as a broadening characteristic. The ratio of the frequency range covered by the broadening characteristic to the target frequency band is used as a coverage metric. The coverage metric is compared with a preset interference assessment threshold. If the coverage metric is lower than the preset low threshold of 0.3, it is determined to be weak interference; if it is higher than the preset high threshold of 0.7, it is determined to be strong interference; and if it is between the two thresholds, it is determined to be moderate interference. The judgment boundaries for each interference level are jointly corrected based on the correlation coefficient and the broadening characteristic in the mapping relationship, resulting in a corrected interference level. Cross-statistics are performed based on the corrected interference level and the energy distribution clustering results to calculate the frequency distribution of each cluster under different interference levels. A correlation matrix between interference intensity and detection performance is constructed. Principal component analysis is used to identify the main interference factors affecting detection and directly apply them to optimize the classification. The output includes interference intensity classification results with three levels: weak, medium, and strong.

[0040] In one implementation, the segmentation of spectral data is achieved using a fixed-step sliding window method. The frequency step size is set to 0.5 Hz. Within the overlapping interval, starting from the initial frequency, a frequency band is divided every 0.5 Hz, with each band having a width of 1 Hz. Statistical analysis is performed on the spectral data within each band. The energy density value is obtained by calculating the average power spectral density of all frequency points within that band. The peak frequency corresponds to the frequency point where the power spectral density is at its maximum within the band. The bandwidth parameter is determined using the half-power point method, i.e., finding the frequency points where the power drops to half the peak value from the peak frequency on both sides; the distance between these two points is the bandwidth. The feature vector composed of these three parameters reflects the energy distribution characteristics of the frequency band. The hierarchical clustering algorithm uses Euclidean distance as a similarity metric when processing the feature vectors. The algorithm starts with each feature vector as an independent cluster and gradually merges the two closest clusters to form a tree structure. After each merge, the within-group variance of the new cluster is calculated, which is the sum of the squared distances from all samples within the cluster to the cluster center. As the number of clusters decreases, the within-group variance shows a trend of first slowly increasing and then rapidly rising. The optimal number of clusters is determined by finding the inflection point of the within-group variance growth rate. In practical applications at shale gas well sites, the optimal number of clusters is usually three to five, corresponding to different types of vibration sources, such as vibration of the main fracturing pump group, vibration of auxiliary equipment, and environmental background noise.

[0041] Specifically, establishing the mapping relationship between vibration intensity and cluster characteristics involves calculating correlation coefficients. For each cluster center, its energy value sequence is extracted and subjected to Pearson correlation analysis with the vibration intensity level sequence for the current time period. The absolute value of the correlation coefficient reflects the degree of association between the cluster and vibration intensity; a positive correlation indicates that the energy of the cluster increases with increasing vibration, while a negative correlation indicates an inverse relationship. By establishing a correlation coefficient matrix, a mapping table from vibration intensity to cluster characteristics is formed, enabling a quantitative description of the variation patterns of spectral characteristics under different vibration conditions.

[0042] Preferably, the power decay rate is calculated using an exponential fitting method. An exponential function is used to fit the distribution of clutter power spectrum broadening in the frequency domain, and the attenuation coefficient of the fitting function is the power decay rate. This rate reflects how quickly clutter energy diffuses from the center frequency to both sides. When the fracturing equipment vibrates violently, the power decay rate is small, and the clutter broadening range is large; when the equipment operates smoothly, the power decay rate is large, and the clutter energy is concentrated in a narrower frequency band. The coverage metric is obtained by calculating the ratio of the frequency range covered by the broadening feature to the frequency band required for target detection. This value is between zero and one; a larger value indicates a more severe impact of clutter on target detection.

[0043] In one possible implementation, the interference assessment threshold is set based on statistical analysis of historical data. Coverage metric data for different fracturing operation stages are collected, their distribution characteristics are statistically analyzed, and the lower quartile of the distribution is set as the low threshold, while the upper quartile is set as the high threshold. This dynamic threshold setting method can adapt to differences in well site environments and equipment configurations. The correction of the interference level judgment boundary by the correlation coefficient is reflected in the fine-tuning of the threshold. When a cluster is highly correlated with vibration intensity, it indicates that the cluster is mainly caused by equipment vibration, and the interference level judgment threshold of the frequency band where the cluster is located is increased accordingly to reduce false judgments.

[0044] For example, the cross-statistical process is implemented by constructing a two-dimensional frequency table. Rows in the table correspond to different clusters, and columns correspond to three interference levels. Each element in the table records the number of times the cluster appears at the corresponding interference level. By calculating the frequency distribution of each row, the interference characteristics of each cluster are obtained. The vibration source cluster is mainly distributed at the strong interference level, while the target signal cluster may span multiple interference levels. The mechanism of principal component analysis in identifying major interference factors is explained. Each row of the correlation matrix represents the interference intensity and detection performance index at a given time. By performing singular value decomposition on the matrix, a set of orthogonal principal component vectors are obtained. The first principal component corresponds to the largest eigenvalue, representing the main direction of data change, and is usually closely related to the operating status of the fracturing pump group. The second principal component may reflect the influence of auxiliary equipment or environmental factors. By analyzing the loading coefficients of the first few principal components, the interference factors with the greatest impact on detection performance are identified. These factors constitute the main basis for interference intensity grading. The output interference intensity grading results not only include three level labels but also a confidence score for each level. The confidence score is obtained by calculating the Mahalanobis distance from the sample to the center of each level; the smaller the distance, the higher the confidence score. This confidence-based grading provides a more precise control basis for subsequent detection threshold adjustments.

[0045] In one embodiment, the interference intensity classification results change significantly as fracturing operations transition from the start-up phase to the stable operation phase. During the start-up phase, as equipment is gradually loaded, the vibration intensity increases, and the interference level gradually transitions from weak to moderate. During the stable operation phase, multiple pump sets operate synchronously, generating strong vibrations that cause most frequency bands to be classified as strong interference levels. By tracking the changes in the interference intensity classification results in real time, a dynamic assessment of the well site safety monitoring environment can be achieved.

[0046] S104. Support vector machine is used to extract boundary features and perform trend fitting on the interference intensity classification results and the spectrum overlap boundary to identify the dynamic threshold adjustment range and determine the detection threshold value after adjustment.

[0047] The interference intensity classification results are converted into numerical feature vectors. Boundary start frequency, termination frequency, and boundary width parameters are extracted from the spectral overlap boundary. A training sample set containing interference level labels and boundary parameters is constructed. A support vector machine is used to classify and train the training sample set, obtaining a decision function that distinguishes boundary features under different interference conditions. Based on the decision function, a distance value sequence from each sample to the classification hyperplane is calculated. This distance value sequence reflects the classification confidence and is physically related to boundary expansion and contraction, as positive distance values ​​indicate an expansion trend and negative values ​​indicate a contraction trend. A polynomial fitting method is used to fit the distance value sequence in chronological order to obtain the slope parameter of the boundary change trend. This slope parameter reflects the expansion or contraction speed of the boundary over time. Based on the slope parameter and the interference intensity classification results at the current moment, a threshold adjustment baseline value is calculated using a piecewise function. A high-weight coefficient a=2 and a low-weight coefficient b=1 are used, defining the baseline value as: absolute slope multiplied by interference level multiplied by weight coefficient. Strong interference uses a, weak interference uses b, and other values ​​are zero. The baseline value is multiplied by a preset unit adjustment amount to obtain the dynamic threshold adjustment amplitude. The original detection threshold value is read from the radar detection parameters. The dynamic threshold adjustment range is added to the original detection threshold value. At the same time, the adjusted value is scaled according to the decision probability of classifying the current sample by the decision function. The decision probability is obtained by the Platt scaling method of SVM. When the probability is higher than 0.8, the original range is maintained. When the probability is between 0.6 and 0.8, the range is reduced by a coefficient of 0.5. When the probability is lower than 0.6, it is reduced to zero. The adjusted detection threshold value is then output.

[0048] In one implementation, the numerical conversion of the interference intensity classification results is achieved using one-hot encoding. The three interference levels—weak, medium, and strong—are mapped to three-dimensional vectors, with weak interference corresponding to a vector of 1-0-0, medium interference to 0-1-0, and strong interference to 0-0-1. During the parameter extraction of the spectral overlap boundary, the boundary start frequency is defined as the lowest frequency point in the overlap interval, and the end frequency is the highest frequency point; the boundary width is calculated by the difference between the two. The encoded interference level vector is concatenated with the three boundary parameters to form a six-dimensional feature vector, which serves as the input to the support vector machine (SVM). The training sample set is constructed by collecting historical data from different fracturing operation stages. Each sample is labeled with its corresponding environmental interference category, such as the equipment startup period, stable operation period, or equipment shutdown period. During training, the SVM uses radial basis functions as kernel functions to map the original feature space to a high-dimensional space, searching for the optimal classification hyperplane in the high-dimensional space. The mathematical form of the decision function includes a combination of support vectors, Lagrange multipliers, and a kernel function; its output value represents the signed distance from the sample point to the classification hyperplane. Positive values ​​indicate that the sample is located on one side of the hyperplane, while negative values ​​indicate that it is located on the other side. The larger the absolute value, the farther the sample is from the hyperplane, and the higher the certainty of the classification. In the application scenario of shale gas well sites, the decision function can identify which typical operating condition the current disturbance pattern belongs to, thus providing a basis for threshold adjustment.

[0049] Specifically, the construction of the distance value sequence involves processing continuous time windows. Each time window corresponds to a feature vector, and a distance value is calculated through a decision function. These distance values ​​are then arranged in chronological order to form a sequence. A third-order polynomial is used for polynomial fitting, and the polynomial coefficients are determined using the least squares method. The fitted curve smooths out random fluctuations in the distance values ​​and extracts the overall trend of boundary changes. The slope parameter is obtained by differentiating the fitted polynomial; the value of the first derivative at the current moment is the slope. A positive slope indicates that the interference boundary is expanding, meaning the range of interference influence is increasing; a negative slope indicates that the boundary is shrinking, and the interference influence is weakening.

[0050] Preferably, the reference value is calculated using a piecewise function. When the slope is positive and the interference level is strong, the reference value is set as the absolute value of the slope multiplied by a high-weighting coefficient; when the slope is negative and the interference level is weak, the reference value is set as the absolute value of the negative slope multiplied by a low-weighting coefficient; in other combinations, the reference value is set to zero. This design ensures that the threshold adjustment direction is consistent with the interference change trend. The unit adjustment amount is preset according to the sensitivity of the radar system, typically ranging from one percent to five percent of the dynamic range of the detection threshold. The dynamic threshold adjustment amplitude is equal to the product of the reference value and the unit adjustment amount; this amplitude can be positive or negative, corresponding to the upward and downward adjustment of the threshold, respectively.

[0051] In one possible implementation, the initial detection threshold is stored in the radar system's configuration parameters, pre-set according to the radar model and detection range. The initial threshold is typically set as a multiple of the background noise power, with this multiple determining the false alarm probability. The dynamic threshold adjustment is then superimposed on the initial threshold to obtain the pre-adjusted threshold. The decision probability is obtained from the support vector machine's probability output, reflecting the confidence level that the current sample belongs to its predicted class. The probability value ranges from zero to one, with values ​​close to one indicating high certainty and values ​​close to 0.5 indicating significant uncertainty.

[0052] For example, the scaling mechanism employs a linear mapping method. When the probability of a decision is higher than 0.8, the adjustment range remains unchanged; when the probability is between 0.6 and 0.8, the adjustment range is multiplied by the probability value to reduce it; when the probability is lower than 0.6, the adjustment range is multiplied by a fixed coefficient of 0.5. This scaling strategy avoids over-adjustment under uncertain classification conditions and improves the stability of threshold adjustment. Threshold adjustment exhibits different characteristics at different stages of fracturing operations. In the initial stage of equipment startup, due to gradually increasing vibration, the threshold value rises in a stepwise manner; during stable operation, the threshold value fluctuates slightly at a high level; after equipment shutdown, the threshold value quickly falls back to the baseline level.

[0053] In one embodiment, when multiple fracturing pump units are alternately started and stopped, the interference environment changes rapidly, and the output value of the decision function frequently crosses zero, indicating that the interference mode is switching between different categories. At this time, the judgment probability is usually low, and the scaling mechanism comes into play, reducing the threshold adjustment range and avoiding drastic fluctuations in unstable environments. The adjusted detection threshold value can adaptively track environmental changes, controlling the false alarm rate while ensuring detection sensitivity.

[0054] S105. The positional distribution of the refined Doppler frequency shift value within the spectral overlap boundary is extracted to evaluate the depth of the target signal embedded in the clutter region and obtain the embedding depth. The intrusion target with low-speed intermittent ground movement is identified to obtain the preliminary target type determination result.

[0055] The distribution of refined Doppler frequency shift values ​​within the spectral overlap boundary is statistically analyzed. The distance between the frequency shift value and the upper and lower limits of the boundary is calculated. The frequency, duration, and interval period of the frequency shift peak are extracted as location distribution features. The product of frequency and period is used as the target activity intensity index I, thus obtaining the spatial distribution parameters of the target signal. Based on the target activity intensity index I in the spatial distribution parameters, the ratio of target signal power Ps to clutter power Pc is calculated, where... k is a preset constant of 1. The signal-to-noise ratio (SNR) is obtained through logarithmic calculation of the ratio. When the SNR is negative, the embedding depth is deep; when the SNR is positive, the embedding depth is shallow. Simultaneously, the target appearance pattern is determined based on the proportion of the duration to the total observation time. The embedding depth is compared with an adjusted detection threshold. If the embedding depth is shallow and the target signal power is higher than the threshold, the target signal is considered separable. The movement characteristics of intermittent ground contact are identified based on the interval period. When the interval period is within a preset range of human walking cycles and exhibits a periodic pattern, it is determined to be a low-speed, intermittent ground contact intrusion target, and a preliminary target type determination result is output.

[0056] In one implementation, the location distribution statistics of refined Doppler frequency shift values ​​are achieved through a time-frequency analysis window. The spectral overlap boundary is divided into multiple sub-intervals, and the frequency shift value occurrences within each sub-interval are counted to form a frequency shift distribution histogram. The upper and lower limits of the boundaries correspond to the lowest and highest frequencies of the overlap interval, respectively, and the distance between the frequency shift value and these two boundaries reflects the relative position of the target signal in the interference environment. When the frequency shift value is close to the lower limit, it indicates that the target signal is at the edge of clutter; when it is close to the upper limit, it has penetrated into the core region of clutter. The calculation of the target activity intensity index comprehensively considers the time domain and frequency domain characteristics. Frequency reflects the activity level of the target within the observation time, and period reflects the regularity of the movement; the product of the two forms a quantified activity intensity value. In the shale gas well site environment, the activity intensity of the actual intruding target usually exhibits a specific range; too high an intensity may indicate equipment vibration interference, while too low an intensity may indicate random noise.

[0057] Specifically, the logarithmic operation of the signal-to-noise ratio (SNR) uses a ten-fold logarithmic form to convert the power ratio into a decibel value. A negative decibel value means that the target signal power is lower than the clutter power, and the signal is overwhelmed by the clutter, corresponding to deep embedding; a positive decibel value means that the target signal is stronger than the clutter, and is only slightly interfered with, corresponding to shallow embedding. The area near zero decibels is the critical state, where the target signal power is comparable to the clutter power.

[0058] Preferably, the criterion for determining the human walking cycle is set to 0.5 to 2 seconds. During normal walking, the interval between each step is approximately 0.5 to 1 second; during slow, creeping movement, the interval can be extended to 1.5 to 2 seconds. When the detected interval falls within this range, and the rate of change over multiple consecutive cycles is less than 20%, it is determined to be a human movement characteristic.

[0059] In one possible implementation, the degree of separability is determined using a dual-condition verification. In addition to comparing the embedding depth with the threshold, the persistence of the target signal must also be verified. If the target signal only exceeds the threshold for a brief moment and cannot form a stable trajectory, it is determined to be non-separable, even if the power is sufficient. Only when the target signal remains detectable for multiple consecutive detection cycles is the determination of an intrusion target output.

[0060] For example, when small equipment intermittently crawls towards the well site, the frequency shift signal generated by its intermittent contact with the ground has obvious periodic characteristics. By identifying this characteristic pattern, accurate determination of low-speed intrusion targets can be achieved.

[0061] S106. Process the preliminary target type determination results of multiple consecutive frames, distinguish the mechanical vibration and intermittent ground contact movement characteristics caused by drilling equipment or fracturing truck operations based on the real-time changes of the current vibration intensity level, and track the motion posture of low-speed targets to determine the stable intrusion identification results.

[0062] The system acquires preliminary target type determination results across multiple consecutive frames, timestamps each frame, and counts the number of consecutive frames with the same determination result. When the number of consecutive frames exceeds a preset frame threshold, the target's position coordinate sequence in each frame is extracted as a temporal feature. The target's movement speed is calculated based on the changes in adjacent positions within the temporal feature. Simultaneously, the system monitors the real-time changes in the current vibration intensity level. If the vibration intensity increases while the target's movement speed suddenly drops to zero, it is determined to be mechanical vibration interference. If the vibration intensity remains low while the target maintains a constant speed, it is marked as a genuine intrusion target. The position coordinate sequence of the genuine intrusion target is smoothed using a Kalman filter, and the changes in movement direction and speed are extracted as motion attitude parameters. When the vibration intensity level is below a preset low-intensity threshold and the motion attitude parameters remain stable within a preset time period, it is confirmed as an intrusion behavior, and a stable intrusion identification result is output.

[0063] In one implementation, processing multiple consecutive frames employs a sliding window mechanism with a window length of ten frames. Each time a new judgment result arrives, the window slides forward one frame. Timestamps, accurate to the millisecond, are used to calculate the inter-frame time interval. The number of identical judgment results within the window is counted; when the number of intrusion target judgment results exceeds 70% of the total number of frames in the window, the judgments are considered consistent. The position coordinate sequence is extracted from the radar detection data of each frame, containing the target's distance and angle information in polar coordinates. The calculation of the target's movement speed takes into account the special environment of shale gas well sites. The change in adjacent positions is calculated using the Euclidean distance formula and then divided by the inter-frame time interval to obtain the instantaneous velocity. When the fracturing pump unit starts, ground vibration causes slight shaking of the radar base; this shaking manifests as synchronous displacement of all target points in target detection. The movement of the actual intrusion target is independent, and its velocity changes are not synchronously affected by equipment vibration.

[0064] Specifically, the correlation between vibration intensity and target characteristics is determined based on statistical correlation. Vibration intensity and target velocity sequences are collected over a period of time, and their Pearson correlation coefficient is calculated. A high correlation indicates that the target motion is dominated by vibration and is a false target; a low or negative correlation indicates that the target has independent motion capabilities. Particularly when vibration intensity increases sharply while the target velocity drops to zero, typically corresponding to the process of a fracturing pump unit starting up from shutdown, the detected target is often clutter caused by equipment vibration.

[0065] Preferably, the state vector of the Kalman filter includes target position and velocity components. The prediction step is based on a uniform motion model, and the update step incorporates new observation positions. The process noise covariance reflects the uncertainty of the target motion, while the observation noise covariance reflects radar measurement errors. Through recursive calculation, the filter outputs a smoothed trajectory, eliminating trajectory jitter caused by random measurement errors.

[0066] In one possible implementation, the extraction of motion attitude parameters involves multiple dimensions. The direction of movement is obtained by calculating the vector angle formed by consecutive position points, and the rate of change of velocity reflects the target's acceleration and deceleration characteristics. When the target maintains uniform linear motion, the attitude parameters are stable; when the target uses a serpentine path to evade, the direction parameters exhibit periodic changes. The low-intensity threshold is set based on the background vibration level, typically twice the vibration intensity of the everyday environment. When the vibration intensity is below this threshold and the standard deviation of the motion attitude parameters is less than a preset value, it indicates that the target is moving stably in a quiet environment, consistent with the behavioral characteristics of a real intruder. In this case, the output stable intrusion identification result has high confidence.

[0067] S107. By combining historical cumulative data of clutter power spectrum broadening with stable intrusion identification results, the current false alarm rate and false alarm rate are evaluated. The dynamic threshold adjustment range is iteratively optimized to obtain the final optimized threshold value, thereby achieving effective identification of low-speed intermittent ground-contact intrusion targets in shale gas well sites.

[0068] The mean and variance of clutter power spectrum broadening at different time periods are extracted from historical cumulative data. The false alarm rate is obtained by dividing the number of false alarm events at each broadening level by the total number of detections. The false detection rate is obtained by dividing the number of missed detections by the number of real targets from stable intrusion identification results. A performance evaluation index including the false alarm rate and the false detection rate is constructed. Based on the performance evaluation index, the gradient descent method is used to calculate the balance point that minimizes the sum of the false alarm rate and the false detection rate. If the false alarm rate exceeds the preset upper limit, the threshold value is adjusted upward; if the false detection rate exceeds the preset upper limit, the threshold value is adjusted downward. The direction and amplitude coefficient of the threshold adjustment are determined. The amplitude coefficient is multiplied by the current threshold value to obtain the threshold increment. The updated threshold value is equal to the current threshold value plus the threshold increment. The updated threshold is used to detect newly acquired data and the performance index is calculated. The iteration stops when the change in the threshold value after three consecutive updates is less than 1% of the threshold value, resulting in the optimized threshold parameters. The optimized threshold parameters are used to detect radar signals. When the signal power exceeds the threshold and the motion characteristics match the low-speed intermittent ground contact mode, it is marked as a suspected target. The stability of the suspected target in the continuous observation period is verified, and the final identification result of the low-speed intermittent ground contact moving intrusion target of the shale gas well field is output.

[0069] In one implementation, historical cumulative data on clutter power spectrum broadening is stored in a circular buffer, with the buffer capacity set to the data volume of the most recent seven days. The mean and variance of the broadening are calculated hourly to form a time series. Broadening levels are categorized into low, medium, and high levels based on variance, with low levels corresponding to stable equipment operation periods and high levels corresponding to peak fracturing operations. The false alarm rate is calculated based on the confusion matrix principle, dividing the number of events identified as intrusions but actually interference by the total number of alarms. The missed detection rate is the proportion of confirmed intrusion events that were not detected by the system. These two indicators together constitute the core parameters of performance evaluation, reflecting the system's detection capability under different operating conditions. The application of gradient descent in threshold optimization requires constructing an objective function. The objective function is defined as the weighted sum of the false alarm rate and the missed detection rate, with the weighting coefficients dynamically adjusted according to the well site's safety level. At higher safety levels, the cost of missed detections is higher; therefore, the weight of the missed detection rate is set to 0.7, and the weight of the false alarm rate is 0.3. Gradient calculation is achieved through numerical differentiation, taking a small increment near the current threshold value to calculate the rate of change of the objective function. The descent direction points in the direction of the fastest decrease in the objective function. The step size controls the convergence speed. The initial step size is set to five percent of the threshold value and gradually decreases as the number of iterations increases.

[0070] Specifically, the direction of threshold adjustment follows these principles: when the false alarm rate exceeds the preset upper limit of 10%, it indicates that the threshold is set too low, and a large number of interferences are mistakenly identified as targets. In this case, the threshold value needs to be increased to reduce false alarms. Conversely, when the false detection rate exceeds the preset upper limit of 5%, it indicates that the threshold is too high, and real targets are missed. In this case, the threshold value needs to be decreased. When both situations occur simultaneously, the priority of adjustment is determined based on weighted importance. The magnitude coefficient is determined by considering the gap between the current performance and the target performance; the larger the gap, the larger the adjustment magnitude, but it cannot exceed 20% of the current threshold value to avoid drastic fluctuations in system performance.

[0071] Preferably, the threshold increment is calculated using an adaptive step-size strategy. In the early stages of iteration, a larger step size is used to quickly approach the optimal region; when the performance improvement slows down, the step size is reduced for fine-tuning. Specifically, the objective function value is monitored for three consecutive iterations; if the decrease is less than one percent, the step size is halved. This strategy ensures both fast convergence and avoids oscillations near the optimum.

[0072] In one possible implementation, the iterative update process employs an online learning model. After each threshold update, the system continues to collect new radar data and statistically analyzes the detection results in real time. The performance metrics of the new data are weighted and averaged with historical metrics, with recent data given higher weight to reflect the timeliness of environmental changes. Convergence judgment considers not only the magnitude of threshold changes but also the stability of performance metrics. The system is considered to have reached a stable state when the standard deviations of the false alarm rate and the missed detection rate are less than preset thresholds for five consecutive iterations.

[0073] For example, different stages of fracturing operations exhibit different optimization characteristics. During the preparation phase, as the equipment starts up, vibration gradually increases, and the threshold value rises in a stepwise manner, with larger adjustments in each iteration. During the stable operation phase, vibration remains high but relatively stable, the threshold value fluctuates within a small range, and the iteration convergence speed is fast. At the end of the operation, the equipment is gradually shut down, vibration weakens, the threshold value decreases smoothly, and the system automatically adapts to environmental changes. The identification of low-speed intermittent ground contact patterns relies on feature matching. The target's movement speed is typically between 0.5 and 2 meters per second, and the ground contact interval exhibits quasi-periodicity, with a coefficient of variation of less than 0.3. The temporal envelope of the signal power is represented as a pulse sequence, with the pulse width corresponding to the ground contact duration, typically between 0.1 and 0.3 seconds. These feature parameters are stored in a feature library and compared with real-time detection signals using a template matching algorithm.

[0074] Stability verification employs a trajectory continuity criterion. A suspected target is detected within five consecutive observation periods, and its positional changes conform to kinematic constraints, meaning the distance between adjacent positions does not exceed the product of the target's maximum speed and the time interval. Targets meeting the continuity criterion are confirmed as genuine intruders, triggering an alarm mechanism.

[0075] In one embodiment, the optimization process in a low-noise nighttime environment exhibits rapid convergence. Due to minimal background interference, both the initial false alarm rate and false negative rate are low, and convergence is achieved after three to five iterations. The optimized threshold value can reliably detect slow-moving intruders while maintaining a low false alarm rate, achieving a detection probability of over 95%. Furthermore, the final identification result includes multi-dimensional information, including target type identification, confidence score, movement trajectory, and threat level assessment. This information is integrated to form structured alarm information, which is pushed to the well site safety management platform, enabling comprehensive monitoring of low-speed, intermittent, ground-contact moving intrusion targets in shale gas well sites.

[0076] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. A radar monitoring and assessment method for the safety status of shale gas well sites, characterized in that, The method includes: The original vibration amplitude data, initial clutter power spectrum information, and target Doppler frequency shift signal are collected, and after filtering, the filtered vibration amplitude data, purified clutter spectrum data, and refined Doppler frequency shift value are obtained. The current vibration intensity level is extracted from the filtered vibration amplitude data. At the same time, the power spectrum broadening of clutter is identified from the cleaned clutter spectrum data. The overlap range is determined by analyzing the spectrum overlap boundary between the low-speed target and the vibration clutter in combination with the refined Doppler frequency shift value. Cluster analysis is performed on the energy distribution characteristics of different frequency bands within the overlapping interval to determine the current vibration intensity level, assess the interference level caused by the clutter power spectrum broadening to target detection, and obtain the interference intensity classification results. Support vector machine is used to extract boundary features and fit trends between the interference intensity classification results and the spectrum overlap boundary, and the dynamic threshold adjustment range is identified to determine the adjusted detection threshold value. Feature extraction is performed on the positional distribution of refined Doppler frequency shift values ​​within the spectral overlap boundary. The embedding depth is obtained by assessing the depth to which the target signal is embedded in the clutter region. Preliminary target type determination results are obtained by identifying intrusive targets that move slowly and intermittently on the ground. Process the preliminary target type determination results of multiple consecutive frames, distinguish the mechanical vibration and intermittent ground contact movement characteristics caused by drilling equipment or fracturing truck operations based on the real-time changes of the current vibration intensity level, and track the motion posture of low-speed targets to determine the stable intrusion identification results. By combining historical cumulative data on clutter power spectrum broadening with stable intrusion identification results, the current false alarm rate and false negative rate are assessed. The dynamic threshold adjustment range is iteratively optimized to obtain the final optimized threshold value, thereby achieving effective identification of low-speed, intermittent, ground-contacting intrusion targets in shale gas well sites.

2. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 1, characterized in that, The process involves acquiring raw vibration amplitude data, initial clutter power spectrum information, and target Doppler frequency shift signal, then filtering to obtain filtered vibration amplitude data, cleaned clutter spectrum data, and refined Doppler frequency shift values, including: Millimeter-wave radar probes are deployed at preset intervals around the perimeter of the shale gas well site. Each probe is equipped with a vibration sensing unit to extract initial clutter power spectrum information from the radar echo and simultaneously capture the original Doppler frequency shift signal generated by target movement to obtain original vibration amplitude data. The original vibration amplitude data is then subjected to low-pass filtering to obtain filtered vibration amplitude data. The initial clutter power spectrum information is then filtered to obtain purified clutter spectrum data. Finally, the original Doppler frequency shift signal is subjected to frequency domain narrowband filtering to extract the effective frequency shift component and obtain the refined Doppler frequency shift value.

3. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 1, characterized in that, The process of extracting the current vibration intensity level based on the filtered vibration amplitude data, identifying the clutter power spectral broadening from the cleaned clutter spectral data, and determining the overlap range by analyzing the spectral overlap boundary between the low-speed target and the vibration clutter using refined Doppler frequency shift values ​​includes: The energy integral value of the vibration signal within the time window is calculated based on the filtered vibration amplitude data and compared with the multi-level vibration threshold to determine the current vibration intensity level. The power spectral density function is extracted from the cleaned clutter spectrum data, and the power spectral frequency width is calculated to obtain the clutter power spectral broadening. A clustering algorithm is used to cluster the overlapping parts by spectral energy density. The number of cluster centers is adjusted according to the current vibration intensity level to identify the overlapping start and end frequencies of the vibration clutter and the target frequency shift, and to determine the overlapping interval range.

4. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 1, characterized in that, The method further includes: acquiring the periodic energy pulse waveform of the pump group's reciprocating motion from the purified clutter spectrum data; acquiring the intermittent frequency shift waveform of the discontinuous ground contact moving target from the refined Doppler frequency shift value; analyzing the dominant position of the pump group's vibration energy pulse in the low-frequency band; evaluating the distribution position of the discontinuous ground contact motion frequency shift waveform in the same low-frequency band; determining the overlap boundary range of the two types of waveforms in the frequency band; and identifying the strength contrast between the pump group's energy pulse and the discontinuous ground contact signal within the overlap area.

5. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 4, characterized in that, The identification of the strength contrast between the pump energy pulse and the intermittent ground contact signal within the overlapping area includes: obtaining the vibration energy envelope curve through short-time Fourier transform to obtain the periodic energy pulse waveform of the pump reciprocating motion; detecting the frequency shift abrupt change generated at the moment the target touches the ground to construct the intermittent frequency shift waveform of the intermittently touching moving target; and comparing the power spectral density values ​​of the periodic energy pulse waveform and the intermittent frequency shift waveform in the same low-frequency band to determine the overlapping boundary and the strength contrast state.

6. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 1, characterized in that, The method involves clustering the energy distribution characteristics of different frequency bands within the overlapping interval to determine the current vibration intensity level, assessing the interference level caused by clutter power spectral broadening to target detection, and obtaining interference intensity classification results, including: The spectral data within the overlapping interval is segmented according to the frequency step size. Energy density, peak frequency, and bandwidth parameters are extracted from each frequency band to form a feature vector. A hierarchical clustering algorithm is used to group these feature vectors, and the within-group variance of different clustering levels is calculated to determine the optimal number of clusters, obtaining energy distribution clustering results for different frequency bands. Based on the energy values ​​of each cluster center in the energy distribution clustering results, a correlation coefficient is calculated between the energy value and the current vibration intensity level, establishing a mapping relationship between vibration intensity and clustering features. The power decrease rate is calculated from the clutter power spectrum broadening as a broadening feature, and the ratio of the frequency range covered by the broadening feature to the target frequency band is used as a coverage metric. The coverage metric is compared with an interference assessment threshold to determine weak, moderate, or strong interference. The judgment boundary for each interference level is corrected based on the correlation coefficient in the mapping relationship, resulting in a corrected interference level. Cross-statistics are performed between the corrected interference level and the energy distribution clustering results to construct a correlation matrix between interference intensity and detection performance. Principal component analysis is used to identify the main interference factors, outputting the interference intensity classification result.

7. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 1, characterized in that, The step of using a support vector machine to extract boundary features and fit trends between the interference intensity classification results and the spectral overlap boundary, and identifying the dynamic threshold adjustment range to determine the adjusted detection threshold value, includes: The interference intensity classification results are converted into numerical feature vectors. Boundary start frequency, termination frequency, and boundary width parameters are extracted from the spectral overlap boundary to construct a training sample set. A support vector machine is used for classification training to obtain a decision function. A distance value sequence is calculated based on the decision function, and trend fitting is performed on the distance value sequence to obtain the slope parameter of the boundary change trend. A baseline value for threshold adjustment is calculated based on the slope parameter and the current interference intensity classification results. The baseline value is multiplied by a unit adjustment amount to obtain the dynamic threshold adjustment amplitude. The original detection threshold value is read from the radar detection parameters. The dynamic threshold adjustment amplitude is added to the original detection threshold value. The adjusted value is scaled according to the decision probability of the decision function, and the adjusted detection threshold value is output.

8. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 1, characterized in that, The process of extracting features from the positional distribution of refined Doppler frequency shift values ​​within the spectral overlap boundary, assessing the depth to which the target signal is embedded in the clutter region to obtain the embedding depth, and identifying intrusion targets moving slowly and intermittently on the ground to obtain preliminary target type determination results includes: The distribution of the refined Doppler frequency shift value within the spectral overlap boundary is statistically analyzed, and the frequency of occurrence, duration, and interval period of the frequency shift peak are extracted as location distribution features to obtain the spatial distribution parameters of the target signal. The ratio of the target signal power to the clutter power is calculated based on the spatial distribution parameters, and the signal-to-noise ratio is obtained by logarithmic operation of the ratio to determine the embedding depth. The target occurrence pattern is determined based on the proportion of duration to total observation time. The embedding depth is compared with the adjusted detection threshold to determine whether the target signal is separable. The intermittent ground contact motion characteristics are identified based on the interval period, and the target is identified as a low-speed, intermittent ground contact intrusion target. The preliminary target type determination result is output.

9. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 1, characterized in that, The process of processing multiple consecutive frames of preliminary target type determination results distinguishes between mechanical vibrations and intermittent ground contact movement characteristics caused by drilling equipment or fracturing truck operations based on real-time changes in the current vibration intensity level, and tracks the motion posture of low-speed targets to determine stable intrusion identification results, including: The system acquires the preliminary target type determination results for multiple consecutive frames, counts the number of consecutive frames with the same determination result, and extracts the target's position coordinate sequence in each frame as a temporal feature. Based on the temporal feature, the system calculates the target's moving speed, monitors the real-time changes in the current vibration intensity level, and determines whether it is a mechanical vibration interference or a real intrusion target. The system performs trajectory smoothing on the position coordinate sequence of the real intrusion target, confirms the intrusion behavior, and outputs a stable intrusion identification result.

10. The radar monitoring and assessment method for the safety status of shale gas well sites according to claim 1, characterized in that, The method combines historical cumulative data of clutter power spectrum broadening with stable intrusion identification results to assess the current false alarm rate and false negative rate, iteratively optimizes the dynamic threshold adjustment range to obtain the final optimized threshold value, and achieves effective identification of low-speed, intermittent, ground-contact moving intrusion targets in shale gas well sites, including: The mean and variance of clutter power spectrum broadening are extracted from historical cumulative data. The false alarm rate is obtained by dividing the number of false alarm events by the total number of detections. The false detection rate is obtained by dividing the number of missed detections by the number of real targets from stable intrusion identification results. A performance evaluation index is constructed. Based on the performance evaluation index, the gradient descent method is used to calculate the balance point that minimizes the sum of the false alarm rate and the false detection rate to update the threshold value and obtain the optimized threshold parameter. The optimized threshold parameter is used to detect radar signals. When the signal power exceeds the threshold and the motion characteristics conform to the low-speed intermittent ground contact mode, it is marked as a suspected target. The stability of the suspected target in the continuous observation period is verified, and the final identification result of the low-speed intermittent ground contact moving intrusion target in the shale gas well field is output.

Images