A large data identification method for periodic pressure of a coal mining face

Abstract

The application discloses a kind of periodic weighting big data identification methods of coal mining face, belong to coal mine strata pressure analysis technical field.The method includes: S1, based on pressure gradient feature cleaning support resistance data, eliminate process noise;S2, construct nonlinear propulsion model, realize the accurate mapping of time domain data to space propulsion degree;S3, working face is divided and is boxed, calculates each regional group energy, constructs multidimensional energy sequence;S4, using the anti-noise trend separation method based on quantile extracts static load baseline, separates load disturbance characteristics, and is enhanced using morphological method;S5, based on robust statistics, construct adaptive threshold, combine geological priori period step distance and carry out consistency check, realize the automatic, accurate identification of periodic weighting.The application solves the problems of traditional method, such as incomplete data cleaning, space-time mapping distortion, weak noise resistance and identification relying on empirical threshold, significantly improves the accuracy, robustness and automation level of weighting identification.

Description

Technical Field

[0001] This invention belongs to the field of coal mine pressure big data analysis technology, specifically involving a method for identifying periodic pressure big data in coal mining faces. By constructing a noise-resistant static load baseline and a periodic load response separation model, the stable identification and quantitative characterization of the periodic pressure process of the roof can be achieved. Background Technology

[0002] During the longwall mining process in coal mines, as the coal mining machine advances, the immediate roof collapses, while the overlying, hard basic roof strata form cantilever beams or masonry beam structures. When the exposed length reaches its limit, the overlying basic roof strata undergo periodic fracture. This process is accompanied by the instantaneous release of enormous elastic energy, which acts violently on the hydraulic supports as dynamic loads, causing the hydraulic support resistance to exhibit periodic pressure characteristics. In recent years, with increasingly depleted resources, coal mining activities have continued to extend deeper, significantly increasing the complexity and stress level of the overlying rock structure, and leading to an upward trend in the frequency of roof collapses. However, traditional methods for identifying periodic pressure based on manual statistics or empirical formulas are insufficient to meet the safety requirements of deep and complex geological conditions. Therefore, how to utilize intelligent monitoring methods to accurately extract and quantify the periodic pressure step distance and pressure intensity of the working face from massive monitoring data is of urgent and practical engineering significance for optimizing support parameters and preventing mine pressure disasters.

[0003] Currently, multi-channel hydraulic support resistance monitoring systems are widely used. However, in actual field operations, support resistance is multi-source and heterogeneous, sampling intervals are inconsistent, sensors may go offline and are susceptible to extreme value interference. At the same time, the progress logs, shift information, and pressure data are not synchronized, making it difficult to form a stable and reusable cycle for automatic pressure identification using traditional methods that rely mainly on single sensors, threshold experience, or manual inspections.

[0004] Chinese patent application CN118760916A discloses a method, electronic device, and medium for analyzing and warning of pressure in fully mechanized mining faces. This method first identifies the entire cutting face of the mining machine and divides it into cycles, then extracts features such as the end-of-cycle resistance. It uses thresholding and clustering to identify pressure-affected areas, and finally calculates the step distance and intensity. It then uses exponential smoothing to predict the next pressure advance and step distance, achieving graded early warning. This method defines the end-of-cycle resistance as the maximum column pressure within a set time period before the hydraulic support lowers its column, and relies on the mining machine to accurately identify the cycle number for pressure identification. However, this method has the following drawbacks in practical applications: the end-of-cycle resistance value is easily distorted by human operations such as premature support movement; if the waveform cutting algorithm misjudges the lowering action, the cycle feature becomes completely invalid, failing to reflect the full-process energy of the roof action; the system relies on external hardware, which often experiences data drift due to vibration or accumulated errors, and hardware failure will interrupt the entire cutting face identification; data filtering methods mainly focus on sampling frequency and amplitude limitations, lacking a deep noise cleaning mechanism, making it easy to mix in non-working resistance data.

[0005] Chinese patent application CN117266936A discloses a method and device for monitoring mine pressure manifestation characteristics based on support resistance. This method filters suspected high-resistance points by setting a threshold, performs spatial clustering on the selected sets within a defined coordinate system, extracts the centroid abscissa of the pressure clusters, and uses the lateral difference between the centroids of adjacent clusters as the pressure step distance, thus obtaining a step distance sequence and visualization results. This patent application uses a coefficient multiplied by the number of times a safety valve is opened as a fixed threshold to calculate the Euclidean distance between point pairs and cluster them to generate pressure regions. The limitations of this method in practical applications are: relying solely on threshold filtering leads to erroneous cleaning of effective process information with a significant increasing resistance trend; and the global clustering logic used in this method is difficult to capture local anomalies, easily causing missed detections. Summary of the Invention

[0006] To address the shortcomings of existing technologies in data cleaning, spatiotemporal mapping, noise resistance, and threshold dependence, this invention provides a method for identifying large data on periodic pressure in coal mining faces that is highly accurate, has strong anti-interference capabilities, does not rely on fixed thresholds, and conforms to the physical laws of roof movement. This method can automatically and accurately identify large data on periodic pressure in coal mining faces.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for identifying periodic pressure data in coal mining faces includes the following steps:

[0009] Step 1: Obtaining and cleaning hydraulic support resistance data;

[0010] High-frequency resistance monitoring data of hydraulic supports in fully mechanized mining faces are obtained. Low resistance data and working resistance data generated by processes including support shifting and column lowering are separated using pressure gradient characteristics. The missing value rate of each measuring point is calculated. The validity of the measuring point data is determined based on the preset missing value rate threshold. The missing value rate of the measuring points is statistically analyzed, invalid measuring points are removed, and spatiotemporal interpolation is performed on the local missing values ​​of valid measuring points to obtain the working resistance data of the cleaned hydraulic supports.

[0011] Step 2, nonlinear spatiotemporal alignment;

[0012] Obtain daily advance data from the working face, construct a nonlinear propulsion model, and use this model to map the cleaned resistance dataset from the time dimension to the spatial propulsion dimension.

[0013] Step 3, spatial energy aggregation in partitioned areas;

[0014] The working face is divided into several independent monitoring areas along the dip direction. Spatial bins are formed with a preset advance step size. The population energy value of each area in each spatial bin is calculated, and a multi-area spatial energy representative sequence is constructed.

[0015] Step 4: Noise-resistant trend separation and morphological feature enhancement;

[0016] The multi-region spatial energy representative sequence is preprocessed by filtering; a noise-resistant trend separation method based on quantile estimation is used to construct a static load evolution baseline, and periodic load perturbation features are separated from the energy representative sequence; the periodic load perturbation features are enhanced using mathematical morphology methods to obtain a periodic load perturbation feature enhanced sequence.

[0017] Step 5: Adaptive threshold and periodic consistency are used to suppress identification;

[0018] Based on the enhanced periodic pressure feature sequence obtained in step 4, an adaptive identification threshold is constructed using robust statistical methods. Combining the periodic pressure step range of geological priors, spatiotemporal logic verification is performed on candidate events that exceed the threshold, and finally, periodic pressure events and their multidimensional state parameters are identified and output.

[0019] Preferably, in step 1, the low-resistance data and working resistance data generated by processes including frame shifting and column lowering are separated using pressure gradient characteristics, specifically including:

[0020] Step 1.1, raw dataset collection;

[0021] The electro-hydraulic control system of the hydraulic support of the longwall mining face is used to acquire the front and rear column resistance monitoring data of the hydraulic support at the set sampling frequency, forming the original pressure-time matrix.

[0022] Step 1.2, Process Feature Identification;

[0023] Calculate the absolute value of the first-order difference of the hydraulic support at adjacent time points;

[0024] If the absolute value of the first-order difference is greater than the preset process judgment gradient threshold, then the support is determined to be in the period of moving or lowering the column during that time period, and the data at that time is process interference data and is marked as invalid.

[0025] If the absolute value of the first-order difference is lower than the preset effective pressure lower limit, the support is determined to be in an unloaded or unconnected state, and the data at that moment is unloaded or unconnected state data and is marked as invalid.

[0026] Step 1.3, Integrity Restoration;

[0027] To prevent information gaps caused by data cleaning, a missing rate threshold R is introduced. m ;

[0028] Calculate the missing rate threshold R for stent pressure data for the current shift. m If R m If R > 0.3, the sensor on the bracket is determined to be faulty, and the data for that bracket is directly discarded. m If the value is less than 0.3, an interpolation algorithm is used to fill in the local null values, with the maximum continuous filling length limited to 5 sampling points.

[0029] Preferably, step 2 includes the following sub-steps:

[0030] Step 2.1: Obtain the advance data for each shift;

[0031] Obtain the daily coal mining shift report for the corresponding date, read the daily footage report, and determine the shift start time, end time, and total footage for the shift. ;

[0032] Step 2.2, Constructing the non-uniform advance function; the non-uniform advance function is constructed using the Sigmoid nonlinear spatiotemporal mapping model. ;

[0033] (1);

[0034] In the formula, x is the normalized time, x∈[0,1]; k is the slope control factor, with a value range of 5~15;

[0035] Step 2.3, Coordinate Mapping; The working resistance dataset cleaned in Step 1 is mapped from the time dimension to the cumulative advance spatial coordinate axis to generate a spatiotemporally aligned dataset; The spatiotemporal coordinate mapping formula is defined as:

[0036] (2);

[0037] In the formula, The cumulative drag progress coordinates corresponding to monitoring time t; This represents the cumulative progress made at the end of the previous coal mining shift. This is the total footage recorded for this shift. It is a non-uniform advance function.

[0038] Preferably, step 3 includes the following sub-steps:

[0039] Step 3.1, space partitioning;

[0040] The advance step size is set at 0.5m, and the cumulative advance of the entire mine is divided into continuous spatial boxes;

[0041] Step 3.2, dividing the work surface into zones;

[0042] The working surface is divided into the upper head region Z. L Central Region Z M and tail area Z R Three regions;

[0043] Step 3.3, Energy Calculation;

[0044] For each box in each region, calculate its population energy. :

[0045] (3);

[0046] in, The group energy value represents the pressure intensity characteristics at the propulsion location; This represents the total number of valid data points within the current spatial container for this region. For the first in the region One effective working resistance value; It is a summation index;

[0047] Step 3.4: Construct a multi-regional spatial energy representation sequence.

[0048] Preferably, step 4 includes the following sub-steps:

[0049] Step 4.1, filter preprocessing;

[0050] Median filtering is applied to multi-region spatial energy representative sequences.

[0051] Step 4.2, based on adaptive noise complete set empirical mode decomposition;

[0052] Based on adaptive noise complete set empirical mode decomposition, multi-scale decomposition is performed on the filtered multi-region spatial energy representative sequence; several intrinsic mode components and a residual term are obtained.

[0053] (4);

[0054] In the formula, The energy representation sequence; For the first One intrinsic modal component; This represents the total number of modal components. The residual term represents the monotonic trend. This is the index of the intrinsic modal components;

[0055] Step 4.3, noise reduction trend separation;

[0056] A noise-resistant trend separation method based on quantile estimation is used to construct a static load evolution baseline. Specifically, it includes;

[0057] Step 4.3.1: For the energy representative sequence E(x), calculate the initial estimate of the low quantile background baseline within a sliding window. ;

[0058] (5);

[0059] In the formula, Here, q is the quantile operator, q is the proportion of the lower quantiles, and L is the sliding window length. To be at the position of advancement The group energy value at that location;

[0060] Step 4.3.2, for Adaptive cascaded Gaussian smoothing is performed to obtain a continuous static load evolution baseline. ;

[0061] Step 4.3.3: Subtract the energy representative sequence E(x) from the baseline B(x) to obtain the periodic load perturbation characteristics. ;

[0062] (6);

[0063] Step 4.4, morphological feature enhancement;

[0064] Feature enhancement is performed using mathematical morphology methods, specifically by performing a top-hat transformation on the periodic load perturbation feature D(x);

[0065] (7);

[0066] In the formula, For erosion calculation; 'b' is the opening operation; 'b' is the structure element. The feature enhancement sequence is periodically compressed.

[0067] Preferably, step 5 includes the following sub-steps:

[0068] Step 5.1, Multi-region feature monitoring and initial threshold judgment;

[0069] Periodic-based pressure feature enhancement sequence An adaptive identification threshold is constructed using robust statistical methods. ;

[0070] (8);

[0071] In the formula, For periodic pressure feature enhancement sequences Median absolute deviation within a local window; This is a robust scaling estimator constructed based on the median absolute deviation; This is the sensitivity adjustment coefficient; This is an absolute safety margin term;

[0072] Step 5.2 involves performing spatiotemporal logic verification by combining the geological prior period with the step range, specifically including:

[0073] The theoretical period that conforms to the movement law of rock strata in the working face is set to press the step distance interval [D] min D max ];

[0074] If the step distance between the candidate pressure event and the previous confirmed pressure event is less than D min If so, it is judged as a false signal and removed;

[0075] If the continuous advance distance exceeds D max If no candidate event is generated, a supplementary search is performed within the out-of-limit interval;

[0076] Step 5.3, Multidimensional feature parameter calculation and output;

[0077] The incoming pressure events verified in step 5.2 are parametrically calculated, and the final output includes 3D feature parameters including periodic incoming pressure location, incoming pressure step distance, incoming pressure intensity level, and incoming pressure area type.

[0078] The beneficial technical effects of this invention are as follows:

[0079] 1) A deep data cleaning mechanism based on process mechanism was constructed, which significantly improved data purity; In view of the shortcomings of existing technology that it is difficult to remove false resistance data such as "frame shifting" and "column lowering" by relying solely on threshold filtering, this invention uses pressure gradient characteristics to accurately identify and separate non-working resistance data generated by human processes.

[0080] 2) Accurate identification of periodic pressure manifestation under non-stationary conditions was achieved; by constructing a noise-resistant background baseline based on rolling quantiles and combining it with smoothing processing, trend separation was achieved under conditions of data dropout, abnormal extreme values ​​and process disturbances, avoiding the problem of traditional mean or minimum value filtering methods being highly sensitive to outliers, and significantly improving the stability and repeatability of periodic pressure manifestation features.

[0081] 3) Construct an adaptive identification mechanism to reduce reliance on manual thresholds: By introducing robust position statistics and robust scale estimation to construct an adaptive identification threshold, this invention can automatically adjust the threshold size according to different propulsion sections and different regional load fluctuation levels, avoiding the problem of fixed experience thresholds failing under complex working conditions and reducing the reliance on manual experience parameters. Attached Figure Description

[0082] Figure 1 This is a schematic diagram of the overall process of the method provided in the embodiments of the present invention.

[0083] Figure 2 This is a thermal diagram of the resistance of the time-series hydraulic support in an embodiment of the present invention.

[0084] Figure 3 This is a thermal diagram of the spatial domain support resistance after spatiotemporal alignment in an embodiment of the present invention.

[0085] Figure 4 This is a schematic diagram of the regional multidimensional spatial domain energy representation sequence constructed in this embodiment of the invention, showing the energy aggregation characteristics of the upper, middle and lower regions.

[0086] Figure 5 This is a schematic diagram of CEEMDAN mode decomposition of spatial domain population energy sequence and extraction of noise-resistant static load baseline in an embodiment of the present invention, showing the mode components from IMF1 to IMF8.

[0087] Figure 6 This is a schematic diagram of the periodic load manifestation feature enhancement sequence obtained after morphological feature enhancement in an embodiment of the present invention.

[0088] Figure 7 This is an example diagram of the periodic pressure identification results of the method of the present invention in an actual fully mechanized mining face, showing the identified pressure position (cumulative advance) and the corresponding pressure step distance. Detailed Implementation

[0089] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0090] To verify the effectiveness of this invention, a fully mechanized mining face in a mine in Huainan, Anhui Province, was selected as the engineering background. This working face has an average burial depth of 850 meters and a dip length of 330 meters, and is equipped with 140 high-extraction, support-type hydraulic supports. The overall flow of the method provided in this embodiment is as follows: Figure 1 As shown, the specific steps include the following;

[0091] Step 1, Data Acquisition and Cleaning: The raw data includes hydraulic support resistance data for the working face collected every 5 minutes and the corresponding daily progress report for the working face. The raw data is cleaned using pressure gradient characteristics to remove non-working resistance noise.

[0092] Obtain hydraulic support resistance monitoring data (sampling frequency 5 minutes) and corresponding daily progress reports for this working face during March 2024. Set the process judgment gradient threshold T. g =5MPa, lower limit of effective pressure T w =10MPa. Iterate through all data. If the absolute value of the first-order difference of resistance ΔP > 5MPa at any given moment, it is determined to be a process interference; if ΔP < 10MPa, it is determined to be an unloading state; all data in the above cases are marked as invalid. Calculate the data missing rate R for each stent. m , for R m Stents with a diameter > 0.3 were rejected, and R was adjusted accordingly. m For supports with a resistance ≤ 0.3, linear interpolation was used to fill in local null values ​​(maximum of 5 consecutive points), ultimately yielding the cleaned effective working resistance dataset. (See attached diagram) Figure 2 )

[0093] Step 2, Nonlinear Spatiotemporal Alignment: A Sigmoid nonlinear spatiotemporal mapping model is introduced. In this embodiment, the slope control factor K=12 is set to accurately map the discrete resistance sequence in the time domain to the cumulative propulsion space domain (see Appendix for alignment details). Figure 3 ).

[0094] Step 3, Zoned Energy Aggregation: The cumulative advance is divided into grid bins with a step size of 0.5m. Based on the energy aggregation characteristics, the working face pressure is aggregated into three regions: upper, middle, and lower. The group energy of each region within each spatial bin is calculated to enhance the characteristics (see attached diagram). Figure 4 ).

[0095] Step 4, Noise Reduction Trend Separation and Feature Enhancement: The CEEMDAN algorithm is used to perform mode decomposition on the energy sequence, decomposing it into 8 intrinsic mode components (IMF1~IMF8) and 1 residual term. By analyzing the average period and energy proportion of each component, IMF1~IMF2 are determined to be high-frequency random noise, IMF3~IMF5 are principal components containing periodic load characteristics, and IMF6~IMF8 and the residual term are low-frequency static load trends. The low-frequency trend term (IMF6~IMF8 + residual) is separated and removed to reconstruct the detrended load sequence. The empirical mode decomposition algorithm is used to adaptively separate and remove the low-frequency trend term from the signal. Then, a mathematical morphological top-hat transform is introduced to perform nonlinear filtering on the detrended signal. A morphological top-hat transform with a structuring element width of b=5 is applied to further smooth out background clutter during non-load periods, significantly highlighting the periodic load pulse signal (corresponding to the attached...). Figure 5 , attached Figure 6 ).

[0096] Step 5, Adaptive Threshold and Periodic Consistency Pressure Identification: Combining the roof geological structure and mining practices of adjacent working faces, the periodic pressure step distance for this working face is defined as 15-25m as a priori information constraint. After processing with the method of this invention, the algorithm successfully identified 5 periodic pressure events within a 100-meter verification section, with pressure step distances of 16.5m, 20.4m, 19.2m, and 21.4m, respectively. Comparing the identification results with the on-site mine pressure observation log, it was found that all identified pressure locations were within the manually counted pressure range, and the pressure step distance identification accuracy was improved to the decimeter level. (Corresponding appendix) Figure 7 ).

[0097] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for identifying periodic pressure data in coal mining faces, characterized in that, Includes the following steps: Step 1: Obtaining and cleaning hydraulic support resistance data; High-frequency resistance monitoring data of hydraulic supports in fully mechanized mining faces are obtained. Low resistance data and working resistance data generated by processes including support shifting and column lowering are separated using pressure gradient characteristics. The missing value rate of each measuring point is calculated. The validity of the measuring point data is determined based on the preset missing value rate threshold. The missing value rate of the measuring points is statistically analyzed, invalid measuring points are removed, and spatiotemporal interpolation is performed on the local missing values ​​of valid measuring points to obtain the working resistance data of the cleaned hydraulic supports. Step 2, nonlinear spatiotemporal alignment; Obtain daily advance data from the working face, construct a nonlinear propulsion model, and use this model to map the cleaned resistance dataset from the time dimension to the spatial propulsion dimension. Step 3, spatial energy aggregation in partitioned areas; The working face is divided into several independent monitoring areas along the dip direction. Spatial bins are formed with a preset advance step size. The population energy value of each area in each spatial bin is calculated, and a multi-area spatial energy representative sequence is constructed. Step 4: Noise-resistant trend separation and morphological feature enhancement; The multi-region spatial energy representation sequence is preprocessed by filtering; A noise-resistant trend separation method based on quantile estimation is used to construct a static load evolution baseline, and periodic load perturbation features are separated from the energy representative sequence. The periodic load perturbation features are enhanced using mathematical morphology methods to obtain a periodic pressure feature enhancement sequence; Step 5: Adaptive threshold and periodic consistency are used to suppress identification; Based on the enhanced periodic pressure feature sequence obtained in step 4, an adaptive identification threshold is constructed using robust statistical methods. Combining the periodic pressure step range of geological priors, spatiotemporal logic verification is performed on candidate events that exceed the threshold, and finally, periodic pressure events and their multidimensional state parameters are identified and output.

2. The method according to claim 1, characterized in that, In step 1, the low-resistance data and working resistance data generated by processes including frame shifting and column lowering are separated using pressure gradient characteristics. Specifically, this includes: Step 1.1, raw dataset collection; The electro-hydraulic control system of the hydraulic support of the longwall mining face is used to acquire the front and rear column resistance monitoring data of the hydraulic support at the set sampling frequency, forming the original pressure-time matrix. Step 1.2, Process Feature Identification; Calculate the absolute value of the first-order difference of the hydraulic support at adjacent time points; If the absolute value of the first-order difference is greater than the preset process judgment gradient threshold, then the support is determined to be in the period of moving or lowering the column during that time period, and the data at that time is process interference data and is marked as invalid. If the absolute value of the first-order difference is lower than the preset effective pressure lower limit, the support is determined to be in an unloaded or unconnected state, and the data at that moment is unloaded or unconnected state data and is marked as invalid. Step 1.3, Integrity Restoration; To prevent information gaps caused by data cleaning, a missing rate threshold R is introduced. m ; Calculate the missing rate threshold R for stent pressure data for the current shift. m If R m If R > 0.3, the sensor on the bracket is determined to be faulty, and the data for that bracket is directly discarded. m If the value is less than 0.3, an interpolation algorithm is used to fill in the local null values, with the maximum continuous filling length limited to 5 sampling points.

3. The method according to claim 1, characterized in that, Step 2 includes the following sub-steps: Step 2.1: Obtain the advance data for each shift; Obtain the daily coal mining shift report for the corresponding date, read the daily footage report, and determine the shift start time, end time, and total footage for the shift. ; Step 2.2, Constructing the non-uniform advance function; the non-uniform advance function is constructed using the Sigmoid nonlinear spatiotemporal mapping model. ; （1）； In the formula, x is the normalized time, x∈[0,1]; k is the slope control factor, with a value range of 5~15; Step 2.3, Coordinate Mapping; The working resistance dataset cleaned in Step 1 is mapped from the time dimension to the cumulative advance spatial coordinate axis to generate a spatiotemporally aligned dataset; The spatiotemporal coordinate mapping formula is defined as: （2）； In the formula, The cumulative drag progress coordinates corresponding to monitoring time t; This represents the cumulative progress made at the end of the previous coal mining shift. This is the total footage recorded for this shift. It is a non-uniform advance function.

4. The method according to claim 1, characterized in that, Step 3 includes the following sub-steps: Step 3.1, space partitioning; The advance step size is set to 0.5m, and the cumulative advance of the entire mine is divided into continuous spatial boxes; Step 3.2, dividing the work surface into zones; The working surface is divided into the upper head region Z. L Central Region Z M and tail area Z R Three areas; Step 3.3, Energy Calculation; For each box in each region, calculate its population energy. : （3）； in, The group energy value represents the pressure intensity characteristics at the propulsion location; This represents the total number of valid data points within the current spatial container for this region. For the first in the region One effective working resistance value; It is a summation index; Step 3.4: Construct a multi-regional spatial energy representation sequence.

5. The method according to claim 1, characterized in that, Step 4 includes the following sub-steps: Step 4.1, filter preprocessing; Median filtering is applied to multi-region spatial energy representative sequences. Step 4.2, based on adaptive noise complete set empirical mode decomposition; Based on adaptive noise complete set empirical mode decomposition, multi-scale decomposition is performed on the filtered multi-region spatial energy representative sequence; Several intrinsic modal components and a residual term are obtained; （4）； In the formula, The energy representation sequence; For the first One intrinsic modal component; This represents the total number of modal components. The residual term represents the monotonic trend. This is the index of the intrinsic modal components; Step 4.3, noise reduction trend separation; A noise-resistant trend separation method based on quantile estimation is used to construct a static load evolution baseline. Specifically, it includes; Step 4.3.1: For the energy representative sequence E(x), calculate the initial estimate of the low quantile background baseline within a sliding window. ; （4）； In the formula, Here, q is the quantile operator, q is the proportion of lower quantiles, and L is the sliding window length. To be at the position of advancement The group energy value at that location; Step 4.3.2, for Adaptive cascaded Gaussian smoothing is performed to obtain a continuous static load evolution baseline. ; Step 4.3.3: Subtract the energy representative sequence E(x) from the baseline B(x) to obtain the periodic load perturbation characteristics. ; （5）； Step 4.4, morphological feature enhancement; Feature enhancement is performed using mathematical morphology methods, specifically by performing a top-hat transformation on the periodic load perturbation feature D(x); （6）； In the formula, For erosion calculation; 'b' is the opening operation; 'b' is the structure element. The feature enhancement sequence is periodically compressed.

6. The method according to claim 1, characterized in that, Step 5 includes the following sub-steps: Step 5.1, Multi-region feature monitoring and initial threshold judgment; Periodic-based pressure feature enhancement sequence An adaptive identification threshold is constructed using robust statistical methods. ; （7）； In the formula, For periodic pressure feature enhancement sequences Median absolute deviation within a local window; This is a robust scaling estimator constructed based on the median absolute deviation; This is the sensitivity adjustment coefficient; This is an absolute safety margin term; Step 5.2, combining the geological prior period to perform spatiotemporal logic verification of the step range, specifically including; The theoretical period that conforms to the movement law of rock strata in the working face is set to press the step distance interval [D] min D max ]; If the step distance between the candidate pressure event and the previous confirmed pressure event is less than D min If so, it is judged as a false signal and removed; If the continuous advance distance exceeds D max If no candidate event is generated, a supplementary search is performed within the out-of-limit interval; Step 5.3, Multidimensional feature parameter calculation and output; The incoming pressure events verified in step 5.2 are parametrically calculated, and the final output includes 3D feature parameters including periodic incoming pressure location, incoming pressure step distance, incoming pressure intensity level, and incoming pressure area type.

Images