Adaptive filtering and compensation method for mine sensor signals and pressure sensor

Abstract

This invention discloses an adaptive filtering and compensation method for mine sensor signals and a pressure sensor, relating to the field of mine monitoring equipment technology. The method includes the following steps: acquiring raw pressure time-series data during the dormant period within a preset time window, and extracting initial noise variance data and initial baseline bias data respectively; constructing an envelope constraint model based on the initial noise variance data; inputting the raw pressure time-series data during the dormant period into the envelope constraint model for deviation point removal processing to generate high-confidence dormant time-series data; and re-extracting new baseline bias data and new noise variance data. This invention performs secondary cleaning on the dormant period data, removing occasional vibration noise, thereby obtaining pure, high-confidence dormant time-series data. This ensures the absolute reliability of the baseline bias and noise variance in all subsequent compensation calculations, effectively avoiding erroneous compensation due to strong environmental interference.

Description

Technical Field

[0001] This invention relates to the field of mining monitoring equipment technology, specifically to an adaptive filtering and compensation method for mining sensor signals and a pressure sensor. Background Technology

[0002] In mine safety monitoring systems, pressure sensors are core components for sensing the underground environment, and their long-term stable operation is crucial for mine disaster early warning. Because the underground working environment is extremely harsh, typically accompanied by high dust, high humidity, and strong mechanical vibration, this places extremely high demands on the sensor's signal acquisition accuracy and anti-interference capabilities.

[0003] To filter out random interference from complex environments and improve data accuracy, existing data processing solutions generally employ specific algorithm models to smooth and denoise the acquired raw signals. In such conventional methods, technicians typically pre-set a fixed set of zero-point references and filter covariance parameters before the equipment is put into use, which serve as the signal calibration standard throughout the entire service life of the equipment.

[0004] Theoretically, under an ideal steady-state model, these static preset parameters can indeed achieve relatively stable error suppression. However, in real-world downhole applications, as the equipment's service life extends, the internal sensitive components inevitably undergo irreversible physical aging under the long-term effects of harsh environmental stress. This underlying physical change leads to dynamic alterations in the equipment's inherent baseline level and measurement noise distribution characteristics. If the static filtering model set at the initial deployment stage is still used, the algorithm system will be unable to perceive the true degree of degradation in the underlying hardware performance, causing a severe disconnect between the preset filtering boundary conditions and the actual distorted physical signal. This not only leads to the gradual failure of conventional denoising functions but may even cause the model output to deviate directionally from the real environmental conditions, resulting in long-term cumulative distortion of the monitoring data. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an adaptive filtering and compensation method for mining sensor signals and a pressure sensor.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] An adaptive filtering and compensation method for mine sensor signals includes the following steps:

[0008] Obtain raw pressure time-series data during the dormancy period within a preset time window, and extract the initial noise variance data and initial baseline bias data respectively;

[0009] An envelope constraint model is constructed based on the initial noise variance data. The original pressure time series data during the dormancy period is input into the envelope constraint model for deviation point removal, generating high-confidence dormancy time series data. New baseline bias data and new noise variance data are then extracted.

[0010] Extract the pre-stored initial baseline data from the factory, and combine it with the new baseline bias data and the new noise variance data to calculate the bias drift rate and variance inflation rate, respectively.

[0011] The bias drift rate and variance inflation rate are cross-weighted and matched to calculate the zero drift compensation weight and measurement noise adaptive adjustment coefficient, respectively. The historical relative working zero parameters and historical measurement noise covariance parameters are then fused to generate the corrected relative working zero parameters and measurement noise covariance parameters.

[0012] The system acquires real-time operating pressure data during the sensor's operation period, and performs compensation processing on the real-time operating pressure data based on the corrected relative operating zero point parameters and measurement noise covariance parameters, outputting the final pressure monitoring data.

[0013] A pressure sensor, comprising:

[0014] Sensor housing;

[0015] The pressure acquisition module is used to sense changes in physical pressure in the mine environment and convert them into raw analog electrical signals for output.

[0016] The analog-to-digital converter module, which is electrically connected to the pressure acquisition module, is used to sample and quantize the raw analog electrical signal and convert it into discrete digital level signals to provide the subsequent stage with raw pressure timing data during the dormant period and real-time operating pressure data.

[0017] The core data processing module, installed inside the sensor housing, is connected to the analog-to-digital conversion module. It contains a microcontroller and a storage medium. The storage medium is used to store computer instruction code and initial factory reference data. The microcontroller is used to execute the steps of any of the methods described in the adaptive filtering and compensation method for mining sensor signals.

[0018] The communication module, installed outside the sensor housing, is connected to the core data processing module and is used to send the final pressure monitoring data output by the core data processing module to the external monitoring system according to the preset mining communication protocol.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] This invention utilizes raw pressure time-series data from the dormant period of a sensor during operation for baseline extraction. Based on the initial noise variance, a dynamic envelope constraint model is constructed to perform secondary cleaning of the dormant data, removing occasional vibration noise. New baseline bias data and new noise variance data representing the pure hardware state are then extracted, resulting in pure, high-confidence dormant time-series data. This ensures the absolute reliability of the baseline bias and noise variance for all subsequent compensation calculations, effectively avoiding erroneous compensation due to strong environmental interference. By using the factory-installed initial baseline data embedded within the chip as a permanent physical anchor point, a longitudinal comparison is made with the currently extracted pure baseline and variance across the entire lifecycle. This transforms invisible hardware degradation into standardized mathematical ratios—bias drift rate and variance expansion rate—achieving precise and objective quantification of the sensor's physical aging depth. By cross-weighting bias drift and variance inflation, smooth compensation weights and adjustment coefficients are generated, which in turn soft-update the historical working zeros and filter noise covariance. This breaks down the physical-mathematical barrier between the underlying hardware attenuation and the upper-level algorithm boundary, enabling the filtering algorithm to automatically adjust its trust in the data as the hardware deteriorates. Furthermore, the smooth fusion mechanism effectively avoids algorithm crashes caused by parameter mutations. Attached Figure Description

[0021] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0022] Figure 1 This is a diagram illustrating the method steps of the present invention;

[0023] Figure 2 This is a flowchart of the present invention;

[0024] Figure 3 This is a front view of the present invention;

[0025] Figure 4 This is a side cross-sectional structural diagram of the present invention.

[0026] The diagram shows: 1. Sensor housing; 2. Pressure acquisition module; 3. Communication module; 4. Core data processing module; 5. Analog-to-digital conversion module. Detailed Implementation

[0027] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0028] Application Overview:

[0029] In the field of mine safety monitoring, especially in the pressure monitoring of underground ventilation and gas pipelines, the accuracy and anti-interference performance of time-series signals are regarded as key indicators for measuring the reliability of safety early warning systems. The generation of such high-fidelity data is essentially a dynamic state optimal estimation process at the signal processing level. That is, through the perception and analog-to-digital conversion of the physical components at the front end of the sensor, the filtering algorithm model is used as the mathematical boundary to effectively separate the real environmental state changes from the complex background noise, thereby outputting an environmental pressure time-series characteristic trajectory with extremely high confidence at the monitoring end.

[0030] However, existing technologies lack a verification mechanism for the time-varying consistency between the degradation of front-end sensor physical hardware and the evolution model of back-end filtering algorithms. This results in the inability to accurately identify baseline concealment shifts and noise variance expansion problems that exist in sensors during long-term service. Baseline concealment shifts manifest as the sensor being able to normally sense pressure fluctuations, but in reality, the operating zero point slowly drifts due to the accumulation of environmental stress. Noise variance expansion manifests as the measurement data dispersion increases and interference intensifies due to the decrease in the internal signal-to-noise ratio caused by the aging of sensitive components. As a result, a strict data correspondence cannot be established between the actual physical state of the sensor and the static boundary conditions preset by the algorithm, causing the system to misjudge or delay feedback on abnormal fluctuations, thereby affecting the accurate extraction of pressure signal dynamic characteristics and the targeted nature of safety warnings.

[0031] For example, in real-world monitoring in coal mines, equipment operates under complex conditions for extended periods. Conventional static filtering models can only smooth short-term random fluctuations, failing to distinguish whether the widening of the fluctuation range is accompanied by a degradation in the sensor's own measurement characteristics. Furthermore, when sensor aging leads to increased measurement noise, the system only records the superficial phenomenon of violent fluctuations, failing to detect the abnormal expansion of background noise variance during dormancy and the decrease in Kalman gain adjustment efficiency. Specifically, the system misjudges baseline offset caused by aging as a genuine gradual pressure change within the pipeline, issuing false alarms, or incorrectly classifies variance expansion as occasional environmental interference and forcibly filters it out. This results in the data processing model continuously solidifying erroneous parameter states, making it impossible to form a high-precision tracking trajectory that conforms to objective physical laws.

[0032] If the above problems are not addressed, the monitoring system will continuously lose its ability to objectively judge the true performance status of sensors under complex operating conditions. Specifically, the failure to identify baseline concealment offsets will cause the algorithm to perform calculations based on erroneous absolute zero points for a long time, resulting in the overall monitoring data deviating from the true environmental pressure benchmark, thereby weakening the system's ability to capture minor pressure anomalies. At the same time, the failure to correct noise variance expansion will cause the core filtering parameters to become misaligned, leading the algorithm to over-rely on degraded measurement data or over-reliance on hysteretic prediction models, ultimately causing the output signal to lose its proper dynamic tracking accuracy. Thus, the inaccuracy of the data processing model will systematically hinder the sensor from achieving adaptive high-precision measurement throughout its entire life cycle, affecting the ultimate achievement of the downhole safety production monitoring goals.

[0033] like Figure 1-4 As shown, an adaptive filtering and compensation method for mine sensor signals includes the following steps:

[0034] Step 1: Acquire raw pressure time-series data during the dormant period within a preset time window, and extract initial noise variance data and initial baseline bias data respectively. Considering the large amount of drastic fluid fluctuations and mechanical interference during mine operations, isolation is first implemented in the time dimension. The dormant period when the sensor is in a non-operational fluid resting state is actively captured, and the underlying physical measurement values ​​within this specific time window are extracted to form the raw pressure time-series data sequence. Subsequently, preliminary statistical analysis is performed on the overall DC level characteristics and AC fluctuation characteristics of this sequence. By calculating the central tendency of the overall data, initial baseline bias data reflecting the current apparent static zero-point offset is extracted. Simultaneously, by evaluating the dispersion of the data from this center, initial noise variance data reflecting the current apparent environmental background noise is extracted. This provides a preliminary reference anchor point for subsequent deep cleaning.

[0035] Step Two: Construct an envelope constraint model based on the initial noise variance data. Input the original pressure time series data during the dormant period into the envelope constraint model for deviation point removal, generating high-confidence dormant time series data. Then, extract new baseline bias data and new noise variance data. Since occasional strong pulse interference such as the start-up and shutdown of heavy mining machinery or rockfalls is unavoidable during the dormant period, directly using the initially extracted parameters would lead to calibration distortion. Therefore, based on the initial noise variance data obtained in Step One, an envelope constraint model with dynamic tolerance upper and lower boundaries is constructed. After inputting the original sequence into this model, the model can automatically identify and remove extreme isolated deviation points that exceed the boundaries of normal physical fluctuations, like a sieve. After cleaning up the occasional vibration noise, the remaining stable sampling points are reassembled in time series to obtain an extremely pure high-confidence dormant time series data. Based on this pure data, feature extraction is performed again to obtain new baseline bias data and new noise variance data that exclude external pulse interference and only represent the current true hardware state of the sensor.

[0036] Step 3: Extract the pre-stored initial factory baseline data, and combine it with the new baseline bias data and new noise variance data to calculate the bias drift rate and variance expansion rate, respectively. To accurately assess the physical aging degree of sensor components, a longitudinal comparison mechanism spanning the entire lifecycle of the device is introduced. The initial factory baseline data and initial factory noise variance data, solidified after calibration in a standard laboratory in a brand-new factory state, are read from the sensor's internal non-volatile storage medium. Using these data as baseline data, the new data extracted in Step 2, representing the current true aging state, are proportionally calculated with the baseline data. The bias drift rate is obtained by calculating the ratio of the current baseline offset to the factory baseline; the variance expansion rate is obtained by calculating the ratio of the current background noise increase to the factory noise. These two ratios transform the sensor's underlying hardware degradation process into standardized numerical indicators that can be directly utilized by the algorithm.

[0037] Step 4: Cross-weighted matching of bias drift rate and variance inflation rate is performed to calculate zero-point drift compensation weight and measurement noise adaptive adjustment coefficient, respectively. Historical relative operating zero-point parameters and historical measurement noise covariance parameters are then fused to generate corrected relative operating zero-point parameters and measurement noise covariance parameters. Considering that hardware aging is a complex process involving multiple coupled factors, this step uses a joint lookup table and cross-weighted calculation of bias drift rate and variance inflation rate to comprehensively assess the current overall performance degradation level. Based on this comprehensive assessment result, two key adjustment factors are calculated: zero-point drift compensation weight for controlling zero-point update confidence, and measurement noise adaptive adjustment coefficient for characterizing noise sensitivity. To prevent drastic oscillations in the monitoring system's state equation caused by parameter mutations, a smoothing mechanism with a flexible transition is adopted. The two adjustment factors are used to weight and scale the historical operating zero-point and historical noise covariance retained from the previous cycle. The final corrected relative operating zero-point parameters and measurement noise covariance parameters not only contain the latest aging state information, but also maintain the consistency of the system's mathematical model.

[0038] Step 5: Acquire real-time operating pressure data during the sensor's operation period, and compensate for the real-time operating pressure data based on the corrected relative zero-point parameters and measurement noise covariance parameters, outputting the final pressure monitoring data. After completing the above adaptive parameter evolution based on the dormancy period, the system enters the normalized operational monitoring phase. For single-point real-time operating pressure data continuously collected by the sensor during the operation period, baseline subtraction is first performed using the corrected relative zero-point parameters from Step 4, thereby completely eliminating the static DC drift error accumulated over long-term use at the physical level, obtaining coarse-adjusted pressure data. Subsequently, the coarse-adjusted data is input into a filter, and the corrected measurement noise covariance parameters from Step 4 are forcibly used as the dynamic boundary condition for the filter's current state estimation. The filter adaptively adjusts the trust weights for the currently drastically fluctuating observation data accordingly, performing dynamic smoothing filtering. Finally, high-precision final pressure monitoring data, eliminating low-frequency aging drift and high-frequency environmental interference, is output to the external monitoring platform.

[0039] Specific Implementation Example 1: Simulating a pressure sensor installed on a gas extraction pipeline in an underground coal mine, which undergoes hardware aging after two years of service and experiences a complete lifecycle process of self-calibration and real-time compensation:

[0040] Background settings and initial state:

[0041] Equipment: Mine gas pipeline pressure sensor;

[0042] Factory health record (stored in internal EEPROM):

[0043] Factory initial baseline data: 100.0 kPa (standard absolute zero).

[0044] Initial noise variance data upon leaving the factory: 0.05 kPa² (extremely low noise floor of brand new components);

[0045] Current physical condition (after two years of service): Due to harsh environment and aging, its actual physical zero point has slowly drifted to about 102.0 kPa, and the background noise has increased.

[0046] Step 1: Dormant Period Data Collection and Coarse Extraction:

[0047] At 2:00 AM, the extraction pump stopped, the airflow in the pipeline stabilized, the sensor determined that it had entered the dormant period, and a 10-second data acquisition window was opened (sampling frequency 100Hz, a total of 1000 points).

[0048] Interference encountered: During these 10 seconds, a heavy electric locomotive happened to pass by in the distant alley, which generated a strong mechanical vibration that lasted for 0.5 seconds, causing the original pressure value to soar to 110.0 kPa.

[0049] Coarse extraction results: Perform overall calculations on these 1000 points.

[0050] Initial baseline bias data: Due to the pull of the high-altitude vibration peak, the arithmetic mean was incorrectly inflated to 102.5 kPa.

[0051] Initial noise variance data: Due to the presence of extreme outliers, the calculated variance distortion is 0.60 kPa². If traditional algorithms directly use these two data points for calibration, severe miscompensation will occur.

[0052] Step 2: Envelope Cleaning and High-Confidence Reconstruction

[0053] To eliminate the interference of locomotive vibration, the envelope constraint model is activated.

[0054] Boundary generation: The global standard deviation is calculated to be approximately 0.77 (the square root of 0.60). Combined with a local sliding window, dynamic upper and lower limits that fluctuate with the data are generated.

[0055] Precise elimination: When the sliding window scanned the vibration data for 0.5 seconds, it found that these points (110.0 kPa) far exceeded the upper limit of the local envelope, and decisively determined them to be occasional vibration noise, and all 50 abnormal points were removed.

[0056] Re-extraction: After concatenating the remaining 950 pure and stable points, recalculate:

[0057] New baseline bias data: falling back to the true 102.1 kPa, representing the current true drift.

[0058] New noise variance data: falling back to the true 0.20 kPa², representing the current aging background noise.

[0059] Step 3: Historical longitudinal benchmarking and quantification of aging:

[0060] Extract the health records from the first day of manufacturing and perform a two-year historical benchmarking calculation with the newly obtained clean data:

[0061] Bias drift rate calculation: (102.1-100.0) / 100.0=2.1%.

[0062] Variance inflation rate calculation: (0.20-0.05) / 0.05=300%.

[0063] These two percentages precisely quantify the sensor's aging: zero-point offset by 2.1% and noise amplification by 3 times.

[0064] Step 4: Cross-weighting and smooth parameter update:

[0065] Using 2.1% and 300% as coordinates, we input them into a two-dimensional aging attenuation mapping matrix for table lookup and performed weighted calculations. Assuming the system table shows relatively severe aging, we set the zero-point drift compensation weight to 0.4 and the noise adaptive adjustment coefficient to 1.5. We retrieved historical parameters saved from the previous period (yesterday): the historical zero point was 101.8 kPa, and the historical noise covariance was 0.15.

[0066] Smooth update at zero point:

[0067] The new baseline (102.1)×0.4 + historical baseline (101.8)×0.6 = 101.92 kPa, which is the corrected relative working zero point, avoiding the oscillation caused by jumping the zero point to 102.1 all at once.

[0068] Noise boundary of adaptive amplifier filter:

[0069] Historical covariance (0.15) × noise adjustment factor (1.5) = 0.225, which is the corrected measurement noise covariance parameter.

[0070] Step 5: Closed-loop compensation during the operation period:

[0071] At 8 a.m., the extraction pump was started, and the operation began, with the pipeline pressure fluctuating wildly.

[0072] Acquiring real-time data at a single point: At a certain moment, the raw reading output by the sensor ADC is 115.00 kPa.

[0073] Step 1: Baseline coarse adjustment (DC removal): Subtract the calculated correction zero point from the reading.

[0074] 115.00-101.92=13.08kPa, which represents the pure airflow change after completely excluding the two-year zero-point drift.

[0075] Step 2: Dynamic Smoothing Filter (AC Removal): The 13.08 kPa value is fed into the Kalman filter. Since the filter has just received amplified noise covariance (0.225, much larger than the factory default of 0.05), the Kalman gain formula will automatically calculate a lower gain value. This means that the filter is aware of the high hardware noise and therefore does not completely trust this drastically fluctuating 13.08 kPa. Instead, it incorporates the smoothed prediction value from the previous moment (e.g., 12.80 kPa) and ultimately outputs a robust 12.85 kPa.

[0076] It can be seen that, according to traditional technical means, the sensor will mistakenly take 100.0 kPa as a permanent zero, causing the output pressure to always be about 2 kPa higher than the actual pressure. If conventional mean filtering is used, the locomotive vibration in the early morning will raise the benchmark to 102.5 kPa, causing the data measured during the day to be lower. If the noise covariance parameter of the filter is not increased, the Kalman filter will still believe in the high-frequency glitches caused by aging, resulting in the output curve being full of jagged edges.

[0077] In the aforementioned technologies, due to the long-term operation of sensors in harsh environments, the internal sensitive elements and conditioning circuits inevitably undergo physical aging. This aging manifests macroscopically as zero-point drift of the signal and an increase in measurement noise variance. Traditional filtering algorithms (such as moving average filtering and conventional Kalman filtering) typically set fixed noise covariance matrices and static zero-point compensation values ​​at the factory. As the usage period extends, these static parameters fail to match the current physical reality of the sensor, leading to compensation failure and continuous accumulation and amplification of measurement errors over time. The core innovation of this invention lies in overcoming the limitations of traditional synchronous noise reduction during operation. It utilizes the original pressure time-series data from the sensor's dormant period during operational use for baseline extraction and constructs a dynamic envelope constraint model based on the initial noise variance. This model performs secondary cleaning of the dormant data, removing occasional vibration noise and re-extracting new baseline bias data and new noise variance data representing the pure hardware state. This results in pure, high-confidence dormant time-series data, ensuring the absolute reliability of the baseline bias and noise variance for all subsequent compensation calculations and effectively avoiding miscompensation caused by strong environmental interference.

[0078] By using the factory-set baseline data embedded inside the chip as an unchanging physical anchor point, and comparing it longitudinally across the lifecycle with the currently extracted clean baseline and variance, the invisible hardware degradation is transformed into standardized mathematical ratios, namely bias drift rate and variance expansion rate, thus achieving accurate and objective quantification of the physical aging depth of the sensor.

[0079] By cross-weighting bias drift and variance inflation, smooth compensation weights and adjustment coefficients are generated, which in turn soft-update the historical working zeros and filter noise covariance. This breaks down the physical-mathematical barrier between the underlying hardware attenuation and the upper-level algorithm boundary, enabling the filtering algorithm to automatically adjust its trust in the data as the hardware deteriorates. Furthermore, the smooth fusion mechanism effectively avoids algorithm crashes caused by parameter mutations.

[0080] During long-term sensor operation, directly extracting denoising parameters during the operational period can severely contaminate the noise characteristic assessment due to fluctuations in the effective pressure signal. This results in the inability to obtain the sensor's true physical degradation baseline, leading to increased true zero-point drift and background noise. Therefore, this paper proposes acquiring the raw pressure time-series data during the dormant period within a preset time window and extracting the initial noise variance data and initial baseline bias data, as detailed below:

[0081] Extract the initial noise variance data and initial baseline bias data separately, including:

[0082] Calculate the arithmetic mean of all discrete sampling points in the raw pressure time series data during the dormancy period to generate the initial baseline bias data;

[0083] The mean of the squared discrete differences between each discrete sampling point in the original pressure time series data during the dormancy period and the initial baseline bias data is calculated to generate the initial noise variance data.

[0084] Among them: preset time window: refers to a continuous data collection interval with a fixed duration that is extracted on the time axis. The length of this interval needs to be long enough to contain a sufficient number of sample points to ensure statistical significance, but it cannot be too long so that the sensor will experience significant slow drift within this interval.

[0085] Raw pressure time series data during dormancy: refers to the sequence of underlying physical measurement values ​​that are continuously collected by pressure sensors at a fixed sampling frequency without any filtering when the mine monitoring system is in a non-operational or relatively quiet state, such as when the relevant large extraction pumps or fans have stopped running and the fluid in the pipeline is in a natural, stable stage without violent fluctuations.

[0086] Initial baseline bias data: refers to the overall center water level or mathematical expectation value of the sensor output signal in the aforementioned dormant state. In an ideal situation with no zero drift, the absolute resting state baseline should be the standard reference zero point; however, in actual aging conditions, this value represents the static zero drift of the sensor at the current moment.

[0087] Initial noise variance data: refers to the degree of dispersion of the raw pressure time-series data during the dormant period around the initial baseline bias data. It reflects the combined intensity of the inherent background electronic noise of the sensor's internal components and the environmental background noise in the mine's resting state during the current operating cycle.

[0088] Step 1: After the system detects that the sensor has entered a sleep period, it opens a preset time window. Assume that within the preset time window, the analog-to-digital conversion module performs discrete sampling at a fixed sampling frequency, acquiring a total of [number missing] samples. The sampling points are arranged in chronological order to generate a discrete data sequence. Let the nth sampling point be... The original pressure value during the dormant period collected at each sampling time was .

[0089] Step 2: To obtain the relative zero-point reference of the sensor in the current state, calculate the arithmetic mean of the data from all discrete sampling points within the entire preset time window. This arithmetic mean represents the DC component of the signal (i.e., the baseline bias). The specific calculation formula is as follows:

[0090] ;

[0091] in: This represents the initial baseline bias data;

[0092] This indicates the total number of discrete sampling points collected within the preset time window;

[0093] Indicates the first The original pressure value during the dormancy period corresponding to each discrete sampling point.

[0094] Step 3: After acquiring the initial baseline offset data reflecting the location of the data center, the system further calculates the degree to which the data at all sampling points deviates from this baseline to quantify the current background noise level. The system calculates the mean of the squared discrete differences between the original pressure values ​​at each discrete sampling point and the initial baseline offset data, thereby generating the initial noise variance data. The specific calculation formula is as follows:

[0095] ;

[0096] in: This represents the initial noise variance data.

[0097] In the aforementioned technology, since the dormancy period avoids strong interference and business signal fluctuations under external operating conditions, it cleverly segments the time dimension to specifically locate and extract the original pressure time series data in the dormant state as the analysis object. The static initial baseline bias data is extracted by the classic arithmetic mean algorithm, and the initial noise variance data reflecting the degree of fluctuation is extracted by the discrete difference square mean algorithm. This provides a pollution-free calibration data source for the entire adaptive compensation system. The baseline and variance data extracted in this way can most realistically reflect the physical aging of the sensor hardware itself, thus laying a highly confident data foundation for the subsequent construction of the envelope constraint model and the calculation of the bias drift rate.

[0098] Although dormant period data avoids strong interference from operational workflows, the complex downhole environment means that occasional falling rocks or distant blasts can still create extreme pulse spikes in the dormant period signal. Directly calculating the mean and variance based on data containing these spikes would skew the baseline due to outliers, resulting in an abnormally large variance and severely misleading subsequent filter parameter updates. Therefore, this paper proposes constructing an envelope constraint model based on initial noise variance data. The original dormant period pressure time series data is input into this model for outlier removal, generating high-confidence dormant time series data. New baseline bias data and new noise variance data are then extracted, as detailed below:

[0099] Although the system was intentionally calibrated during a relatively static dormancy period, unpredictable strong physical pulse interference (such as transient shock waves from blasting) is always present in the harsh mine environment. Directly mixing these towering spikes in the data into the overall sequence to calculate the baseline zero point and noise variance would severely skew the calculated average and amplify the calculated variance, rendering subsequent aging assessments completely ineffective. Therefore, this paper proposes constructing an envelope constraint model based on the initial noise variance data. The original pressure time-series data during the dormancy period is input into the envelope constraint model for deviation point removal, generating high-confidence dormancy time-series data, specifically including:

[0100] Global fluctuation baseline data is extracted based on initial noise variance data;

[0101] Based on the time series characteristics of global fluctuation benchmark data and dormant period original pressure time series data, an upper envelope point data sequence and a lower envelope point data sequence are generated, and an envelope line constraint model is formed by the upper envelope point data sequence and the lower envelope point data sequence.

[0102] Traverse the original pressure time series data during the dormancy period and determine whether the data of each sampling point is between the upper limit point data and the lower limit point data of the envelope at the same moment;

[0103] If it is determined that the data of a certain sampling point exceeds the range formed by the upper limit point data and the lower limit point data of the envelope at the corresponding time, the sampling point data is marked as occasional vibration noise data and is removed.

[0104] The remaining unremoved sampling points are concatenated according to time sequence to generate high-confidence dormant time series data.

[0105] A dynamic envelope constraint decision mechanism based on statistical principles was introduced. By extracting a global fluctuation benchmark and combining it with the local features of data fluctuations over time, the algorithm delineates a closely fitting dynamic tolerance channel for the data waveform, much like laying tracks. Through point-by-point traversal and logical comparison, precise removal of isolated pulse spikes was achieved, followed by seamless timing stitching. This effectively and thoroughly blocks external mechanical stress interference from contaminating the underlying physical and electrical parameter calibration process. Not only does it filter out highly destructive extreme value interference, but more importantly, it preserves the small and authentic background noise fluctuation characteristics in the original data to the greatest extent possible. The reconstructed high-confidence dormant time-series data truly achieves the goal of removing falsehoods and retaining the true information, providing absolutely reliable underlying data support for the subsequent accurate extraction of various parameters reflecting the physical aging of sensor hardware.

[0106] Even during dormancy, natural airflow within mine pipelines (such as weak gasping flows caused by the pressure difference between the inside and outside of the tunnel) leads to localized time-varying characteristics in background noise, sometimes smooth and sometimes slightly fluctuating. If a fixed-width envelope model is used (e.g., setting a fixed value for the mean fluctuation as the boundary throughout), the envelope will be too wide in areas of smooth airflow (failing to remove subtle abrupt changes), while it will be too narrow in areas of slightly larger natural airflow fluctuations (mistakenly treating normal gasping flows as noise and removing them), thus compromising the objectivity of subsequent historical feature comparisons. Therefore, this paper proposes a dynamically updated boundary in the envelope constraint model, specifically including:

[0107] The initial noise variance data is squared to obtain the global initial standard deviation data, and the global initial standard deviation data is used as the global fluctuation benchmark data.

[0108] Set a sliding data window with a preset step size on the raw pressure time series data during the dormancy period;

[0109] Local time-series data within a sliding data window are acquired over time, and the local mean and local standard deviation of the local time-series data are calculated respectively.

[0110] The ratio of local standard deviation data to global initial standard deviation data is calculated to generate a scaling factor that is updated in real time with time steps;

[0111] Calculate the product of the scaling factor and the global initial standard deviation data, perform point-by-point calculations with the local mean data for the product result, and generate the upper envelope point data sequence and the lower envelope point data sequence in time sequence.

[0112] By introducing a mechanism that dynamically compares local features with a global benchmark using a sliding window, a dimensionless scaling factor is cleverly constructed by dividing the microscopic local standard deviation by the macroscopic global standard deviation. This factor is used to adjust the product result in real time and combined with the local mean to construct a dynamic boundary algorithm with adaptive tracking characteristics. This gives the envelope constraint model high elasticity and adaptability. When the sliding window scans an area with extremely smooth data, the local standard deviation decreases, the scaling factor is less than 1, and the envelope automatically tightens, thus accurately capturing and eliminating minor abnormal spikes. When scanning an area with slightly larger natural airflow fluctuations, the scaling factor increases, and the envelope automatically widens to accommodate these reasonable physical fluctuations without false positives. This flow-shaped dynamic update strategy effectively improves the accuracy and robustness of abnormal deviation point elimination.

[0113] Among them: Envelope constraint model: a mathematical model with dynamic upper and lower boundaries, consisting of upper limit point data sequences and lower limit point data sequences that conform to the data fluctuation trend, used to define the reasonable physical fluctuation range of the sensor during the dormant period.

[0114] Deviation points (incidental vibration noise data): These refer to data points that exceed the upper and lower boundaries of the envelope constraint model in time. These points are usually not the steady-state background noise generated by the internal components of the sensor, but rather instantaneous, high-intensity mechanical vibration or impact interference caused by external physical events such as blasting in mines or sudden start-up and shutdown of heavy machinery.

[0115] High-confidence dormant time series data: After removing the deviation points from the original time series data, the remaining pure sampling points within a reasonable fluctuation range are reassembled in the original time sequence to form a new data sequence.

[0116] The new baseline bias data is an arithmetic mean recalculated based on high-confidence dormant time-series data. It eliminates the pull of extreme interference points and more accurately represents the sensor's current true zero-point drift state.

[0117] New noise variance data: The variance is recalculated based on high-confidence dormant time-series data. It eliminates the amplification effect of occasional impulse noise and more realistically reflects the inherent background noise level of the sensor after the current hardware aging.

[0118] Step 1: First, perform a square root operation using the extracted initial noise variance data to obtain the global initial standard deviation data representing the macroscopic fluctuation amplitude of the entire dormancy period:

[0119] ;

[0120] in: This represents the global fluctuation benchmark data (i.e., the global initial standard deviation data).

[0121] Step 2: To enable the envelope to adapt to local fluctuations in the signal, a value containing [a specific element] is set on the original pressure time series data during the dormant period. A sliding data window with sampling points. As the window slides along the time axis at preset steps, local time-series data within the window are extracted, and the local mean and local standard deviation are calculated.

[0122] For the One sliding window position:

[0123] ;

[0124] ;

[0125] in: Indicates the first Local mean data within a sliding window;

[0126] This represents the fixed amount of data (window width) in a sliding data window.

[0127] Indicates the first The first sliding window Local time-series data (i.e., the corresponding raw pressure values);

[0128] Indicates the first Local standard deviation data within a sliding window.

[0129] Step 3: Calculate the ratio of the local standard deviation to the global initial standard deviation to generate a scaling factor reflecting the severity of local fluctuations. Then, calculate the product of this scaling factor and the global initial standard deviation, and use this product as a margin to perform addition and subtraction operations with the local mean, thereby dynamically generating the upper and lower limits of the envelope. Specifically:

[0130] ;

[0131] ;

[0132] ;

[0133] ;

[0134] in:

[0135] Indicates the first The scaling factor that updates in real time with each sliding window over time;

[0136] This represents the product of the scaling factor and the global initial standard deviation data (i.e., dynamic volatility margin).

[0137] This represents the upper envelope point data generated at the corresponding time.

[0138] This represents the lower bound point data of the envelope generated at the corresponding time.

[0139] Step 4: Traverse the original pressure time series data during the dormancy period and determine the original sampling point. Is it between its corresponding time? and Between. If or If a point is identified as sporadic vibration noise data caused by an external impact, it is determined to be such data and is directly removed from the sequence. The remaining discrete sampling points that meet the constraints are then concatenated in their original time order to form a sequence with a total length of [length missing]. High-confidence dormant time-series data sequence (assuming its internal elements are ) ).

[0140] Step 5: For the high-confidence dormant time-series data generated after cleaning, extract more accurate parameters:

[0141] ;

[0142] ;

[0143] in: This indicates the newly extracted baseline bias data;

[0144] This represents the total number of sampling points for high-confidence dormant time-series data;

[0145] This indicates the first [number] in a high-confidence dormant time-series data sequence. Data from one unremoved sampling point;

[0146] This indicates the newly extracted noise variance data.

[0147] The aforementioned technology introduces a dynamic envelope constraint model based on the global-local fluctuation ratio. By using a sliding data window to perceive the local fluctuation state of the data in real time, a scaling factor is dynamically generated using the local and global standard deviations. This allows the upper and lower boundaries for anomaly detection to adaptively widen or tighten with fluctuations in the local base current. This boundary is then used to precisely remove isolated spike pulses, and finally, secondary parameter extraction is performed on the cleaned sequence. This effectively avoids the drawbacks of using a fixed threshold, which might lead to the accidental deletion of normal fluctuations or the omission of small spikes. Because the contamination from occasional vibration interference is eliminated, the newly recalculated baseline bias data and noise variance data achieve extremely high confidence levels. They are free from environmental artifacts and purely characterize the physical state of the sensor hardware itself, i.e., the true attenuation level and inherent noise, providing the most solid data support for achieving adaptive and accurate compensation throughout the entire lifecycle.

[0148] Because denoising under long-term harsh operating conditions only relies on the baseline and noise variance extracted at the current moment, it only reveals the data characteristics under the current state, but cannot determine how much the sensor has deteriorated relative to its healthy state. The lack of historical references throughout the entire lifecycle makes it impossible to accurately assess the aging depth of the sensor's physical components, and consequently, to calculate a scientifically reasonable error compensation weight, easily leading to blind or insufficient compensation. Therefore, this paper proposes extracting pre-stored initial factory baseline data, combining it with new baseline bias data and new noise variance data, and calculating the bias drift rate and variance expansion rate, including:

[0149] Analyze the initial factory baseline data and extract the initial factory baseline data and initial factory noise variance data;

[0150] Calculate the first difference between the new baseline bias data and the initial baseline data from the factory, and calculate the ratio of the first difference to the initial baseline data from the factory to obtain the bias drift rate;

[0151] Calculate the second difference between the new noise variance data and the initial noise variance data from the factory, and calculate the ratio of the second difference to the initial noise variance data from the factory to obtain the variance inflation rate.

[0152] Among them, the pre-stored initial reference data refers to the reference physical parameters that are fixed and written into the internal storage medium (such as non-volatile memory EEPROM or Flash) when the pressure sensor is initially calibrated under a standard controlled environment before being put into use. It represents the perfect reference characteristics of the sensor in a brand new state without any aging or wear, and specifically includes the initial baseline data and the initial noise variance data.

[0153] Bias drift rate: A dimensionless ratio used to quantify the degree of shift of the current static zero point of a sensor from its brand-new state since it was put into use, due to physical aging of the internal sensing element, material fatigue, or long-term environmental stress.

[0154] Variance inflation rate: A dimensionless ratio used to quantify the extent to which the inherent background measurement noise dispersion of a sensor increases relative to its brand-new state during its service life due to physical attenuation factors such as internal circuit aging and signal-to-noise ratio decline.

[0155] Step 1: The core data processing module inside the sensor accesses a specific address range of its storage medium, parses and extracts two data items that are permanently stored in the device in its brand-new factory state: the initial baseline data representing the perfect zero-point reference state and the initial noise variance data representing the ideal background noise level.

[0156] Step 2: Extract new baseline bias data representing the current true state of the sensor. Then, calculate the absolute difference between this data and the initial factory baseline data, i.e., the first difference, which represents the absolute physical quantity of baseline drift. To obtain a uniform attenuation ratio, the system divides this first difference by the initial factory baseline data to obtain the bias drift rate. The specific calculation formula is as follows:

[0157] ;

[0158] in: This represents the calculated bias drift rate;

[0159] This represents new baseline bias data re-extracted based on high-confidence sequences;

[0160] This represents the initial baseline data extracted from the storage medium at the factory.

[0161] Step 3: Similarly, extract new noise variance data representing the sensor's current inherent noise level. Calculate the absolute difference between this data and the initial factory noise variance data, i.e., the second difference, which represents the absolute physical increment of the increase in background noise. Subsequently, calculate the ratio of this second difference to the initial factory noise variance data to obtain the variance inflation rate. The specific calculation formula is as follows:

[0162] ;

[0163] in: This represents the calculated variance inflation rate;

[0164] This represents new noise variance data re-extracted based on a high-confidence sequence;

[0165] This represents the initial factory noise variance data parsed and extracted from the storage medium.

[0166] The aforementioned technology incorporates a longitudinal benchmarking mechanism across the entire device lifecycle by introducing a time span into the algorithm. The perfect static benchmark, embedded in the chip at the factory, serves as an invariant reference anchor point. Difference and ratio calculations are performed between this benchmark and the extracted current real data, eliminating the dimensional limitations of purely absolute physical quantities and transforming them into standardized bias drift rate and variance expansion rate, reflecting the proportion of degradation. This allows the physical degradation and aging process of the sensor to be accurately quantified by a mathematical model. Through the drift rate and expansion rate, the lifecycle evolution trajectory of sensor hardware performance degradation over time is precisely depicted. These two parameters serve as a bridge connecting the underlying physical changes in the hardware with the upper-level digital filtering algorithm, providing a crucial quantitative driving basis for subsequent adaptive cross-calculation of compensation weights and dynamic adjustment of filter boundaries.

[0167] After obtaining the sensor's performance degradation indicators, directly replacing the old baseline with a newly extracted baseline, or arbitrarily adjusting the filter noise parameters, can easily cause severe oscillations in the system's state equation, i.e., excessive parameter mutations. This introduces new mathematical noise into the compensation process itself, disrupting the continuity and smoothness of the monitoring data. Therefore, this paper proposes a cross-weighted matching of the bias drift rate and variance inflation rate to calculate the zero-point drift compensation weight and the measurement noise adaptive adjustment coefficient, respectively. Furthermore, it fuses historical relative operating zero-point parameters and historical measurement noise covariance parameters to generate corrected relative operating zero-point parameters and measurement noise covariance parameters, as detailed below:

[0168] After obtaining the two degradation indices of bias drift and variance expansion from the sensor, directly substituting these physical ratios as control parameters into the filter model would lead to two serious consequences: First, it ignores the physical coupling amplification effect between the two aging phenomena; second, the physical attenuation rate may exhibit short-term jumps or large absolute values, and directly using it to update the filter model would cause the state matrix to diverge, leading to mathematical collapse at the system level. Therefore, this paper proposes a cross-weighted matching of the bias drift rate and variance expansion rate, calculating the zero-point drift compensation weight and the measurement noise adaptive adjustment coefficient, including:

[0169] Construct a two-dimensional aging decay mapping matrix, and input the bias drift rate and variance inflation rate as retrieval coordinates into the two-dimensional aging decay mapping matrix;

[0170] Obtain the matrix element values ​​corresponding to the search coordinates in the matrix, and use them as multidimensional aging coupling feature data;

[0171] The multidimensional aging coupling feature data is input into the first normalized smoothing function to generate a value range of... Zero-point drift compensation weight;

[0172] The multidimensional aging coupling feature data is multiplied with a preset noise sensitivity amplification factor to generate an adaptive adjustment coefficient for measurement noise.

[0173] A dual safety mechanism of lookup table dimensionality reduction and function smoothing is introduced. First, a pre-trained two-dimensional aging decay mapping matrix is ​​used to transform two independent physical ratio coordinates into a single feature quantity that comprehensively considers the coupling effect. Then, a normalized smoothing function is used to soften this feature quantity and constrain it within a safe range to generate zero-point fusion weights. A sensitive scaling factor is used to linearly transform this feature quantity into a scaling adjustment coefficient for the noise covariance. This effectively eliminates the impedance mismatch between the underlying physical features and the upper-level mathematical algorithm. The two-dimensional matrix lookup table mechanism greatly simplifies the computational complexity of the end-side microcontroller while ensuring the scientific nature of the coupling evaluation. The introduction of the first normalized smoothing function effectively ensures that the generated weights are always within the algorithm's controllable safe convergence domain. This allows the subsequent real-time compensation module to maintain absolute stability in mathematical operations even when facing extreme aging conditions where the sensor is on the verge of damage, avoiding a drop or gap in safety monitoring data due to sudden parameter changes.

[0174] During long-term adaptive calibration of sensors, due to extreme and unpredictable external environmental factors or transient disturbances in internal circuits, even after cleaning, the newly extracted baseline or noise parameters may occasionally deviate from the historical evolution trajectory. If the system uses hard updates (i.e., directly replacing the historical relative operating zero-point parameters with new baseline bias data), it will cause an artificial step-like abrupt change (step response) in the sensor's monitoring curve after each sleep period; while abrupt changes in the noise covariance will cause severe oscillations in the Kalman filter gain, and may even lead to the filter model diverging and failing. To address this, a method is proposed to fuse the historical relative operating zero-point parameters and historical measurement noise covariance parameters separately to generate corrected relative operating zero-point parameters and measurement noise covariance parameters, including:

[0175] The first bias component data is obtained by multiplying the new baseline bias data with the zero-point drift compensation weight.

[0176] The second bias component data is obtained by multiplying the historical relative working zero-point parameters and the inverse weights. The inverse weight data is obtained by subtracting the zero-point drift compensation weight from the constant.

[0177] The first bias component data and the second bias component data are weighted and fused to generate the corrected relative working zero point parameters;

[0178] The product of the adaptive adjustment coefficient of measurement noise and the historical measurement noise covariance parameter is calculated to generate the corrected measurement noise covariance parameter.

[0179] By introducing a weighted fusion (smooth transition) algorithm based on complementary weights to update the zero point, and a scaling algorithm based on adaptive coefficients to update the covariance, a first-order low-pass filter is essentially constructed in the time domain through the allocation of zero-point drift compensation weights and inverse weight data. It dynamically determines whether the system should more trust and closely approximate the new state, or more conservatively maintain the historical state, using the previously evaluated aging comprehensive weights. This achieves flexible evolution of the sensor's core operating parameters. The weighted fusion mechanism can absorb abnormal jumps within individual calibration cycles, ensuring that the system's operating baseline shifts slowly, smoothly, and seamlessly along the sensor's actual physical aging trajectory. This guarantees the absolute stability of the underlying mathematical state transition equations of the adaptive filter and ensures that the final pressure monitoring data sent to the external monitoring system does not produce any artificial mathematical discontinuities across the entire time axis, effectively improving signal fidelity and system robustness.

[0180] Among them: Two-dimensional aging attenuation mapping matrix: a pre-fixed data lookup table matrix, whose horizontal and vertical axes are the bias drift rate and variance expansion rate, respectively. The element values ​​stored inside reflect the overall aging coupling degree of the sensor under different drift and expansion combinations.

[0181] The data structure of the two-dimensional aging decay mapping matrix is ​​specifically represented as a lookup table pre-installed in the microcontroller's storage medium. To balance the microcontroller's storage space and lookup table accuracy, the system divides the bias drift rate and variance inflation rate into multiple discrete interval levels.

[0182] The following is a typical example Numerical embodiment of the two-dimensional aging degradation mapping matrix. In this embodiment, the row coordinates represent the degradation level of variance inflation rate, the column coordinates represent the degradation level of bias drift rate, and the intersecting values ​​within the table are the output multidimensional aging coupling feature data:

[0183]

[0184] Multidimensional aging coupled characteristic data: By inputting the current bias drift rate and variance expansion rate as coordinates, the specific values ​​retrieved and output from the data lookup table matrix represent the comprehensive degradation index of the sensor at the current moment due to the combined effects of zero-point offset and noise amplification.

[0185] Zero-point drift compensation weight: a value range in The coefficients between these coefficients are used to determine the system's confidence in the newly extracted baseline bias data and the retention rate of historical working zero parameters when updating the working zero point.

[0186] Measurement noise adaptive adjustment factor: A dynamic multiplier used to amplify or reduce the measurement noise covariance of the previous cycle based on the aging of the sensor, thereby adjusting the filter's confidence in the observed data in real time.

[0187] Historical relative working zero-point parameter: the baseline zero-point reference value determined at the end of the previous observation cycle or the previous dormancy period.

[0188] Historical measurement noise covariance parameter: The parameter of the observation noise covariance matrix used by the adaptive filter (such as the Kalman filter) in the previous observation period.

[0189] Corrected relative operating zero-point parameters or measurement noise covariance parameters: After a comprehensive evaluation of the current aging state, the latest zero-point reference values ​​and filter noise parameters are generated, which will serve as the foundation for real-time filter compensation in the next stage (operation period).

[0190] Step 1: The aging of physical hardware is often not a deterioration of a single metric; there is a non-linear physical coupling between zero-point drift and noise amplification. An internal two-dimensional aging attenuation mapping matrix is ​​constructed, using bias drift rate and variance inflation rate as input retrieval coordinates (indices). Multi-dimensional aging coupling feature data is extracted from the matrix:

[0191] ;

[0192] in: This represents the multidimensional aging coupling feature data obtained by looking up a table;

[0193] Represents the retrieval function for a two-dimensional aging decay mapping matrix;

[0194] Indicates the bias drift rate;

[0195] This represents the variance inflation rate.

[0196] Step 2: After obtaining the coupling features, calculate two independent coefficients used for zero-point correction and noise correction, respectively. First, input the multidimensional aging coupling feature data into a preset first normalization smoothing function (such as a variation of the Sigmoid function) to generate values ​​within... Zero-point drift compensation weights for the interval. Simultaneously, the coupled feature data is simply multiplied by a preset noise sensitivity amplification factor to generate adaptive adjustment coefficients for measurement noise.

[0197] ;

[0198] ;

[0199] in: This represents the generated zero-point drift compensation weight;

[0200] This represents the preset first normalized smoothing function;

[0201] This represents the adaptive adjustment coefficient for the generated measurement noise;

[0202] This represents the preset noise sensitivity amplification factor (a constant used to adjust the sensitivity of noise amplification).

[0203] In a preferred embodiment of the present invention, the first normalized smoothing function specifically adopts an exponential saturation function or a modified sigmoid function. Taking the exponential saturation function as an example, its specific mathematical expression is as follows:

[0204] ;

[0205] Represents the right to zero-point drift compensation;

[0206] Represents the preset smoothing adjustment coefficient;

[0207] This represents multidimensional aging coupling characteristic data.

[0208] When the sensor is brand new and not yet aged ,but Calculate This indicates a lack of trust in the new baseline, insisting on preserving the historical perfect zero point entirely. As the sensor ages, getting bigger and bigger It will gradually approach 0, thus making A smooth convergence to 1 indicates increasing trust in the newly extracted baseline. Regardless of aging severity, the weights will never equal or exceed 1, ensuring the absolute safety of the fusion algorithm. It completely abandons the traditional linear mapping method, effectively avoiding divergence in the algorithm's state equation caused by excessively large feature data due to extreme sensor aging, leading to output weights exceeding normal limits. This function strictly converges the fusion weights to... Within the range, not only is the mathematical rigor of weight allocation guaranteed, but also the efficient and flexible transformation of underlying physical aging characteristics into filter adjustment parameters is realized, thereby improving the anti-oscillation capability of the adaptive compensation process.

[0209] Step 3: To avoid drastic baseline jumps due to anomalies, a weighted smoothing fusion method is used to update the zero point. The product of the baseline bias data and the zero-point drift compensation weight is calculated to obtain the first bias component data; simultaneously, the product of the historical relative working zero-point parameters and the inverse weight (1 minus the zero-point drift compensation weight) is calculated to obtain the second bias component data. Finally, the two are added together to complete the fusion.

[0210] ;

[0211] ;

[0212] ;

[0213] ;

[0214] in: This represents the calculated first bias component data;

[0215] This indicates the extracted new baseline bias data;

[0216] Indicates the zero-point drift compensation weight;

[0217] Indicates inverse weighted data;

[0218] This represents the calculated second bias component data;

[0219] This represents the historical relative working zero-point parameters stored in the system;

[0220] This represents the final generated and updated corrected relative working zero-point parameter.

[0221] Step 4: Finally, the calculated adjustment coefficients are used to adaptively update the noise boundary conditions of the filter. The adjustment coefficients are directly multiplied by the historical measured noise covariance parameters to obtain the covariance of the current period.

[0222] ;

[0223] in: This represents the generated and updated corrected measurement noise covariance parameter;

[0224] This represents the adaptive adjustment coefficient for measurement noise;

[0225] This represents the stored historical measurement noise covariance parameter.

[0226] The aforementioned technology employs a two-dimensional mapping matrix to nonlinearly cross-couple and reduce the dimensionality of multi-dimensional aging features, followed by a soft update mechanism using weighted fusion and coefficient scaling. When updating zero points, absolute replacement is not used; instead, the weights of historical and new data are calculated. When updating noise, the historical covariance is adaptively scaled based on the coupling features. This strikes a balance between tracking the actual aging of the sensor and maintaining the stability of the system's mathematical model. On one hand, the cross-weighted matrix ensures the scientific and comprehensive nature of physical parameter updates; on the other hand, weighted fusion achieves a smooth transition of the sensor's core operating parameters. This flexible update mechanism effectively prevents the filtering algorithm from collapsing due to parameter mutations, ensuring that the corrected boundary conditions can smoothly and reliably drive the subsequent real-time Kalman filtering algorithm.

[0227] Traditional mine monitoring sensors face two major types of interference during operation: first, accumulated DC interference (zero-point drift); and second, high-frequency AC interference (measurement noise) from machine noise. Conventional algorithms either smooth high-frequency noise while ignoring the overall rise or fall of the baseline, or use a constant filter gain. This leads to the filter blindly trusting contaminated observation data even when the sensor is severely aged and has extremely high background noise, resulting in severely distorted final monitoring curves. Therefore, this paper proposes acquiring real-time operating pressure data during sensor operation and compensating for this data based on corrected relative zero-point parameters and measurement noise covariance parameters, outputting the final pressure monitoring data as follows:

[0228] Traditional digital filtering algorithms are ineffective when dealing with data incorporating aging errors. Direct filtering can only remove high-frequency jitter, but the overall data still deviates from the true value due to zero-point drift. Furthermore, if a Kalman filter with fixed parameters is used, when the sensor is severely aged and the signal-to-noise ratio drops sharply, the fixed boundary conditions can cause the filter to mistakenly believe the observed data is still healthy, thus assigning it excessively high gain weights. This results in a jagged and severely distorted pressure curve in the final output. Therefore, this paper proposes a method to compensate for real-time operating pressure data based on corrected relative operating zero-point parameters and measurement noise covariance parameters, outputting the final pressure monitoring data, including:

[0229] Acquire real-time operating pressure data at a single point, incremented by the sampling clock.

[0230] Subtract the corrected relative operating zero-point parameter from the single-point real-time operating pressure data to eliminate the static baseline drift characteristics accumulated over the usage cycle and generate coarse-adjusted pressure data.

[0231] The coarse-adjusted pressure data is input into a preset adaptive filter model. The corrected measurement noise covariance parameter is used as the boundary condition for updating the current state of the adaptive filter model. The coarse-adjusted pressure data is dynamically smoothed and filtered to generate and output the final pressure monitoring data.

[0232] A decoupled dual-compensation architecture is employed. The first layer is arithmetic coarse adjustment, which uses high-confidence relative zeros for subtraction to directly smooth out static drift characteristics accumulated over the filter's lifespan. The second layer is dynamic fine adjustment, breaking the filter's black box by injecting a corrected measurement noise covariance parameter, incorporating physical aging characteristics, into the denominator of the filter gain equation. The autoregressive properties of the Kalman filter are used to smooth the coarse-adjusted data mathematically. Due to the introduction of dynamic boundary conditions, when hardware aging leads to increased actual measurement noise, the denominator of the gain formula increases, and the calculated gain automatically decreases. This means that when faced with poor-quality data, the filter intelligently lowers its confidence in the current observation reading and relies more on the system's smooth historical predictions. This ensures that regardless of the sensor's service life or the degree of aging of its internal components, the final pressure monitoring data output to the external monitoring system remains smooth, stable, and with an absolutely accurate macroscopic baseline.

[0233] In complex mine environments, while baseline coarse-tuned pressure data eliminates macroscopic DC bias, it remains riddled with high-frequency random AC interference caused by mechanical vibration, airflow turbulence, and circuit thermal noise. Directly outputting the coarse-tuned data results in a screen full of sawtooth waves for the monitoring system, making it impossible to accurately set alarm thresholds. Using simple moving average filtering leads to severe system lag, failing to capture genuine gas or airflow surges in a timely manner. Therefore, this paper proposes inputting the coarse-tuned pressure data into a pre-defined adaptive filter model for dynamic smoothing filtering, including:

[0234] The adaptive filter model calls upon the prior error covariance data from the previous period and the corrected measurement noise covariance parameters to calculate the Kalman gain characteristic data for the current observation period.

[0235] Based on Kalman gain characteristic data, coarse adjustment pressure data, and predicted observation characteristic data, the posterior state estimation data is calculated.

[0236] The posterior state estimation data is output as the final pressure monitoring data after eliminating interference from external complex environmental noise.

[0237] By inputting baseline-drift-removed data into a standard Kalman filter state machine, a classic recursive framework of prediction-measurement-correction is employed. The rigid, unchanging constant matrix in the classical formula is replaced by a dynamically corrected measurement noise covariance parameter. Data fusion is achieved through correction via the product of physical residuals and dynamic gain. This achieves both physical consistency and mathematically optimal smoothness of the filter. Because the corrected measurement noise covariance parameter includes prior aging information, when the sensor is severely aged and measurements are extremely unreliable, the denominator increases, and the calculated dynamic gain automatically decreases. The algorithm automatically switches to: distrusting the current noisy, coarse-tuned pressure data and closely following the smoothed predicted observational feature data; conversely, if the sensor is healthy, it keenly captures real-time changes. This mechanism effectively eliminates interference from complex external environmental noise without increasing hardware costs, balancing the smoothness of signal filtering with real-time response to actual pressure changes.

[0238] Among them: Sensor operating period: In contrast to the dormant period, this refers to the period when the mine monitoring system is in normal working condition, large equipment (such as extraction pumps and fans) is running, and the ambient airflow and pipeline pressure are experiencing real and drastic fluctuations. During this period, the sensors are in a state of continuous real-time monitoring of external physical pressure.

[0239] Real-time operating pressure data (single-point real-time operating pressure data): During the sensor's operating period, the analog-to-digital conversion module continuously collects and outputs the original digital signal of the current moment, which includes the actual environmental pressure, superimposed with the system baseline drift, and environmental composite noise, according to the sampling clock frequency set by the system.

[0240] Coarse-tuned pressure data: This is the preliminary data obtained by subtracting the system's static baseline drift characteristics (i.e., relative operating zero-point parameters) from the single-point real-time operating pressure data. This data has eliminated aging DC bias errors but still contains high-frequency measurement noise.

[0241] Adaptive filter model: In this application, it specifically refers to a Kalman filter model in which the boundary conditions (observation noise covariance) are dynamically updated with the aging degree of the sensor, which is used to perform dynamic smoothing filtering on coarse-tuned pressure data.

[0242] Kalman gain characteristic data: The core dynamically adjusted weight parameter in the adaptive filter model determines whether the system trusts the observed real-time coarse-adjustment pressure data more or the model's own prediction data more at the current moment.

[0243] Predictive observational feature data: The filter calculates the theoretically expected pressure value at the current moment based on the system state and physical state transition equations of the previous moment.

[0244] Prior error covariance data: a statistical assessment of the uncertainty of the filter's predicted state before incorporating current observation data.

[0245] Posterior state estimation data (final pressure monitoring data): The filter calculates the optimal state estimate that is closest to the actual situation at the current moment after fusing the predicted value and the current observation value, which is used as the final output.

[0246] Step 1: Under normal operating conditions, obtain the current time as the sampling clock advances. The single-point real-time operating pressure data is obtained. First, to eliminate the static zero-point drift accumulated over many years of use, the corrected relative operating zero-point parameter is directly subtracted from this real-time data. The specific calculation formula is as follows:

[0247] ;

[0248] in: Indicates in Coarse-scale pressure data generated in real time;

[0249] Indicates in Real-time operational pressure data for a single point is acquired at all times;

[0250] This represents the corrected relative operating zero-point parameter.

[0251] Step 2: Input the coarse-tuned pressure data (with baseline bias removed) into the adaptive filter model. Before performing filter fusion, the filter calls the prior error covariance data from the previous period and forces the corrected measurement noise covariance parameter as the boundary condition of the current observation equation to calculate the Kalman gain characteristic data for the current observation period. The specific calculation formula is as follows:

[0252] ;

[0253] in: Indicates the calculated result Kalman gain characteristic data at time step;

[0254] Indicates system call The prior error covariance data at time step (derived from the previous time step);

[0255] This represents the corrected measurement noise covariance parameter (which replaces the fixed measurement noise matrix in traditional Kalman filtering).

[0256] Step 3: After calculating the Kalman gain, use this gain to weight and fuse the observed coarse-tuned pressure with the theoretically predicted pressure to obtain the posterior state estimation data. The specific calculation formula is as follows:

[0257] ;

[0258] in: This represents the calculated posterior state estimation data, which is the final pressure monitoring data output to the outside after eliminating interference from external complex environmental noise.

[0259] This indicates that the system calculates based on the previous state. Predictive observational feature data at any given time;

[0260] This represents the Kalman gain characteristic data calculated above;

[0261] express Coarse-adjusted pressure data at any given time (as the current actual observation).

[0262] Subsequently, based on Update the error covariance for use in the next period, i.e.:

[0263] , and so on.

[0264] It should be noted that the adaptive filter model is a state machine based on an autoregressive algorithm using Kalman filtering. The complete dynamic smoothing filtering process includes two core steps: state prediction and state update. Before calculating the Kalman gain (i.e., state update) using the corrected measurement noise covariance parameter, the model first performs a state prediction operation, which includes the following sub-steps:

[0265] Calculating and predicting observational characteristic data: Based on the posterior state estimation data of the previous observation period (i.e., the final pressure monitoring data output in the previous period) and a preset physical state transition matrix, the model extrapolates the theoretical pressure prediction value for the current moment. It is assumed that the pressure changes in the mine pipeline over a short period conform to a stationary Markov gradual change model, and its state transition equation is:

[0266] ;

[0267] in, Indicates the current Predictive observational feature data at any given time;

[0268] express Posterior state estimation data generated at time 1;

[0269] This is the system state transition constant (typically set for steady airflow models). );

[0270] To control the input matrix (usually set to 0);

[0271] It is an external control variable.

[0272] Calculating the prior error covariance data: Based on the posterior error covariance data of the previous observation period, and combined with the system's own process noise, the model estimates the uncertainty of the current predicted state. Its prediction error covariance equation is:

[0273] ;

[0274] in, Indicates the derived current Prior error covariance data at time points;

[0275] express The posterior error covariance data at time 1;

[0276] This is the transpose of the state transition constant;

[0277] The process noise covariance parameter is preset for the system, representing the minimal uncertainty inherent in the system's theoretical mathematical transfer model itself.

[0278] Only after completing the above state prediction steps and obtaining and Subsequently, the adaptive filter model further uses the corrected measurement noise covariance parameter to replace the constant observation noise matrix in the standard formula, and then calculates the Kalman gain. And complete the final fusion output.

[0279] In the aforementioned technique, by performing absolute arithmetic subtraction (coarse adjustment) before entering the complex filter, the static drift characteristics are thoroughly stripped away using the high-confidence zero points extracted in the early stage. In the filtering stage, the rigid setting of fixed observation noise covariance in traditional Kalman filtering is broken, and dynamic noise parameters, which incorporate variance inflation rate, are forcibly injected into the gain formula. This ensures that regardless of how many years the sensor ages, the macroscopic baseline of its output signal remains firmly anchored at the absolute physical zero point. The adaptive Kalman filtering effect is extremely significant: when the sensor is severely aged, the gain formula makes... By automatically reducing the size of the data, the algorithm will reduce its reliance on the current coarse observation data and instead rely more on the model's smooth predictions, thus still being able to output smooth and accurate final pressure monitoring data even under harsh working conditions.

[0280] A pressure sensor, comprising:

[0281] Sensor housing 1; Sensor housing 1 is the external protective and mechanical support structure for the entire pressure sensor. Considering the harsh working conditions often present in coal mines, such as high concentrations of methane, coal dust, and water spray, sensor housing 1 is preferably made of high-strength, corrosion-resistant stainless steel or special alloy materials, and is designed for sealing according to intrinsically safe or explosion-proof standards for mining applications, such as achieving IP65 or IP68 protection levels. The housing not only withstands external mechanical impacts but also effectively shields against interference from external strong electromagnetic equipment, providing a stable physical working environment for the internal sensitive components.

[0282] Pressure acquisition module 2 is used to sense changes in physical pressure in the mine environment and convert them into raw analog electrical signals. This module typically contains a high-precision pressure-sensitive core, such as a piezoresistive, capacitive, or resonant micro-mechanical silicon strain gauge. When environmental physical pressure acts on the sensitive core, the resistance or capacitance of its internal bridge undergoes microscopic deformation, generating a weak millivolt-level analog voltage or current signal. Furthermore, this module can integrate front-end analog signal conditioning circuitry to perform preliminary low-noise amplification and temperature compensation for the weak signal.

[0283] The analog-to-digital converter module 5, electrically connected to the pressure acquisition module 2, samples and quantizes the raw analog electrical signal, converting it into discrete digital level signals to provide subsequent stages with raw pressure timing data during the dormant period and real-time operating pressure data. The core function of this module is to perform high-frequency sampling and quantization of the aforementioned irregular raw analog electrical signal, converting it into discrete digital level signals recognizable by the microcontroller. To match the requirements of the aforementioned adaptive filtering algorithm for capturing minute noise characteristics, a high-resolution (e.g., 24-bit or higher) Sigma-Delta analog-to-digital converter is preferred. This module continuously provides the digitized underlying data stream to subsequent stages according to the system's preset sampling clock frequency. In the system's resting state, it provides raw pressure timing data during the dormant period; in the operational state, it provides real-time operating pressure data.

[0284] The core data processing module 4, installed inside the sensor housing 1, communicates with the analog-to-digital converter module 5. It contains a microcontroller and a storage medium. The storage medium stores computer instruction code and initial factory reference data. The microcontroller executes steps of methods such as adaptive filtering and compensation for mining sensor signals. The core data processing module 4, mounted on the main circuit board inside the sensor housing 1, is the edge computing brain of the entire sensor. It communicates with the analog-to-digital converter module 5 via an internal bus (such as SPI, I2C, etc.). This module highly integrates an industrial-grade microcontroller (MCU, such as an ARM Cortex-M series chip) and non-volatile storage media (such as EEPROM or Flash memory).

[0285] Storage medium: Used to persistently store computer instruction code (i.e., the code implementation of all the aforementioned adaptive filtering and compensation algorithms), and a dedicated secure read-only area is provided for permanently storing the initial factory reference data (initial factory baseline data and initial factory noise variance data) generated by the device in the calibration laboratory, as a reference anchor point throughout the entire life cycle.

[0286] Microcontroller: Used to call the code in the storage medium and execute the core computational logic in the aforementioned method embodiments. The microcontroller can perform resting self-test, envelope cleaning, and aging cross-weighted evaluation during the equipment's sleep period; and during the operation period, it uses the corrected zero point and dynamically updated Kalman gain to perform end-side closed-loop compensation calculations on real-time operating pressure data.

[0287] The communication module 3, installed outside the sensor housing 1, connects to the core data processing module 4 and is used to send the final pressure monitoring data output by the core data processing module 4 to an external monitoring system according to a preset mining communication protocol. The communication module 3 acts as a gateway interface for data transmission; it is primarily installed outside the sensor housing 1 (or the antenna / interface terminal extends outside the housing), and internally connects to the core data processing module 4. This module is used to send the final pressure monitoring data output by the core data processing module 4 after adaptive filtering, noise reduction, and elimination of aging drift to an external underground substation or surface monitoring system according to a preset mining communication protocol.

[0288] Depending on the mine's networking requirements, communication module 3 can be a wired RS485 bus module, CAN bus module, or mining industrial Ethernet module; or it can be a wireless communication module such as LoRa, NB-IoT, or WiFi-6. Regardless of the hardware physical layer used, this module incorporates a standard data frame encapsulation protocol that conforms to the "General Technical Requirements for Coal Mine Safety Monitoring Systems," ensuring that high-fidelity pressure monitoring data can be integrated into the mine's overall IoT system in real time and stably.

[0289] By physically integrating high-precision sensing hardware (acquisition module 2 and conversion module 5) with a data processing hub (core data processing module 4) equipped with edge computing capabilities, the traditional lag mode of mining sensors only handling data acquisition and remote host handling processing is broken. This edge-side adaptive compensation architecture enables the equipment to independently complete a perfect closed loop from physical sensing and error self-healing to digital output underground, effectively reducing the bandwidth pressure on communication links and the computational burden on external systems.

[0290] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method of adaptive filtering and compensation of a mine sensor signal, characterized in that, Includes the following steps: Obtain raw pressure time-series data during the dormancy period within a preset time window, and extract the initial noise variance data and initial baseline bias data respectively; An envelope constraint model is constructed based on the initial noise variance data. The original pressure time series data during the dormancy period is input into the envelope constraint model for deviation point removal, generating high-confidence dormancy time series data. New baseline bias data and new noise variance data are then extracted. Extract the pre-stored initial baseline data from the factory, and combine it with the new baseline bias data and the new noise variance data to calculate the bias drift rate and variance inflation rate, respectively. The bias drift rate and variance inflation rate are cross-weighted and matched to calculate the zero drift compensation weight and measurement noise adaptive adjustment coefficient, respectively. The historical relative working zero parameters and historical measurement noise covariance parameters are then fused to generate the corrected relative working zero parameters and measurement noise covariance parameters. The system acquires real-time operating pressure data during the sensor's operation period, and performs compensation processing on the real-time operating pressure data based on the corrected relative operating zero point parameters and measurement noise covariance parameters, outputting the final pressure monitoring data.

2. The adaptive filtering and compensation method for mining sensor signals according to claim 1, characterized in that: An envelope constraint model is constructed based on the initial noise variance data. The original pressure time-series data during the dormancy period is then input into the envelope constraint model for deviation point removal, generating high-confidence dormancy time-series data, specifically including: Global fluctuation baseline data is extracted based on initial noise variance data; Based on the time series characteristics of global fluctuation benchmark data and dormant period original pressure time series data, an upper envelope point data sequence and a lower envelope point data sequence are generated, and an envelope line constraint model is formed by the upper envelope point data sequence and the lower envelope point data sequence. Traverse the original pressure time series data during the dormancy period and determine whether the data of each sampling point is between the upper limit point data and the lower limit point data of the envelope at the same moment; If it is determined that the data of a certain sampling point exceeds the range formed by the upper limit point data and the lower limit point data of the envelope at the corresponding time, the sampling point data is marked as occasional vibration noise data and is removed. The remaining unremoved sampling points are concatenated according to time sequence to generate high-confidence dormant time series data.

3. The adaptive filtering and compensation method for mining sensor signals according to claim 2, characterized in that: The boundaries in the envelope constraint model are dynamically updated, specifically including: The initial noise variance data is squared to obtain the global initial standard deviation data, and the global initial standard deviation data is used as the global fluctuation benchmark data. Set a sliding data window with a preset step size on the raw pressure time series data during the dormancy period; Local time-series data within a sliding data window are acquired over time, and the local mean and local standard deviation of the local time-series data are calculated respectively. The ratio of local standard deviation data to global initial standard deviation data is calculated to generate a scaling factor that is updated in real time with time steps; Calculate the product of the scaling factor and the global initial standard deviation data, perform point-by-point calculations with the local mean data for the product result, and generate the upper envelope point data sequence and the lower envelope point data sequence in time sequence.

4. The adaptive filtering and compensation method for mining sensor signals according to claim 1, characterized in that: Extract the initial noise variance data and initial baseline bias data separately, including: Calculate the arithmetic mean of all discrete sampling points in the raw pressure time series data during the dormancy period to generate the initial baseline bias data; The mean of the squared discrete differences between each discrete sampling point in the original pressure time series data during the dormancy period and the initial baseline bias data is calculated to generate the initial noise variance data.

5. The adaptive filtering and compensation method for mining sensor signals according to claim 1, characterized in that: Extract the pre-stored initial factory baseline data, and combine it with the new baseline bias data and new noise variance data to calculate the bias drift rate and variance inflation rate, including: Analyze the initial factory baseline data and extract the initial factory baseline data and initial factory noise variance data; Calculate the first difference between the new baseline bias data and the initial baseline data from the factory, and calculate the ratio of the first difference to the initial baseline data from the factory to obtain the bias drift rate; Calculate the second difference between the new noise variance data and the initial noise variance data from the factory, and calculate the ratio of the second difference to the initial noise variance data from the factory to obtain the variance inflation rate.

6. The adaptive filtering and compensation method for mining sensor signals according to claim 1, characterized in that: The bias drift rate and variance inflation rate are cross-weighted and matched to calculate the zero-point drift compensation weight and measurement noise adaptive adjustment coefficient, including: Construct a two-dimensional aging decay mapping matrix, and input the bias drift rate and variance inflation rate as retrieval coordinates into the two-dimensional aging decay mapping matrix; Obtain the matrix element values ​​corresponding to the search coordinates in the matrix, and use them as multidimensional aging coupling feature data; The multidimensional aging coupling feature data is input into the first normalized smoothing function to generate a value range of... Zero-point drift compensation weight; The multidimensional aging coupling feature data is multiplied with a preset noise sensitivity amplification factor to generate an adaptive adjustment coefficient for measurement noise.

7. The adaptive filtering and compensation method for mine sensor signals according to claim 6, characterized in that: The historical relative operating zero-point parameters and historical measurement noise covariance parameters are fused separately to generate corrected relative operating zero-point parameters and measurement noise covariance parameters, including: The first bias component data is obtained by multiplying the new baseline bias data with the zero-point drift compensation weight. The product of the historical relative working zero point parameter and the reverse weight is calculated to obtain the second bias component data, wherein the reverse weight data is obtained by subtracting the zero point drift compensation weight from the constant 1. The first bias component data and the second bias component data are weighted and fused to generate the corrected relative working zero point parameters; The product of the adaptive adjustment coefficient of measurement noise and the historical measurement noise covariance parameter is calculated to generate the corrected measurement noise covariance parameter.

8. The adaptive filtering and compensation method for mining sensor signals according to claim 1, characterized in that: The real-time operating pressure data is compensated based on the corrected relative operating zero-point parameters and measurement noise covariance parameters, and the final pressure monitoring data is output, including: Acquire real-time operating pressure data at a single point, incremented by the sampling clock. Subtract the corrected relative operating zero-point parameter from the single-point real-time operating pressure data to eliminate the static baseline drift characteristics accumulated over the usage cycle and generate coarse-adjusted pressure data. The coarse-adjusted pressure data is input into a preset adaptive filter model. The corrected measurement noise covariance parameter is used as the boundary condition for updating the current state of the adaptive filter model. The coarse-adjusted pressure data is dynamically smoothed and filtered to generate and output the final pressure monitoring data.

9. The adaptive filtering and compensation method for mine sensor signals according to claim 8, characterized in that: The coarse-adjusted pressure data is input into a preset adaptive filter model for dynamic smoothing filtering, including: The adaptive filter model calls upon the prior error covariance data from the previous period and the corrected measurement noise covariance parameters to calculate the Kalman gain characteristic data for the current observation period. Based on Kalman gain characteristic data, coarse adjustment pressure data, and predicted observation characteristic data, the posterior state estimation data is calculated. The posterior state estimation data is output as the final pressure monitoring data after eliminating interference from external complex environmental noise.

10. A pressure sensor, characterized in that, include: Sensor housing; The pressure acquisition module is used to sense changes in physical pressure in the mine environment and convert them into raw analog electrical signals for output. The analog-to-digital converter module, which is electrically connected to the pressure acquisition module, is used to sample and quantize the raw analog electrical signal and convert it into discrete digital level signals to provide the subsequent stage with raw pressure timing data during the dormant period and real-time operating pressure data. The core data processing module, installed inside the sensor housing, is communicatively connected to the analog-to-digital conversion module. It contains a microcontroller and a storage medium. The storage medium is used to store computer instruction code and initial factory reference data. The microcontroller is used to execute the steps of the method as described in any one of claims 1 to 9. The communication module, installed outside the sensor housing, is connected to the core data processing module and is used to send the final pressure monitoring data output by the core data processing module to the external monitoring system according to the preset mining communication protocol.

Images