Mine production safety early warning method and system

Abstract

The present application relates to the technical field of production safety monitoring, and discloses a mine production safety early warning method and system, the method comprising collecting and cleaning multi-source environment original data, smoothing and missing value completion to obtain a standard data set; screening multiple risk superposition marks through data consistency verification, calculating risk mark values according to preset weights and dynamically improving sampling frequency; carrying out fluctuation analysis on fine data to update weights to obtain a comprehensive risk assessment score; generating an early warning signal when the score is higher than an alarm threshold and calibrating sensor deviation, adjusting collection density and power consumption distribution according to real-time early warning results, and determining an optimized multi-source information collection mode. The present application realizes multi-source data fusion and accurate identification of multiple risks, has high real-time and accuracy of early warning, low false and missed alarm rates, and reasonable power consumption distribution, and is suitable for intelligent safety early warning in complex underground working conditions.

Description

Technical Field

[0001] This invention relates to the field of production safety monitoring technology, and in particular to a method and system for early warning of mine production safety. Background Technology

[0002] Currently, with the gradual application of intelligent sensing, the Internet of Things, and big data analytics technologies in the mining industry, underground environmental monitoring based on artificial intelligence and big data analysis has become a crucial technical support for ensuring operational safety. Due to the complex and ever-changing underground mining environment, with real-time fluctuations in environmental parameters and the potential for multiple coupled risks to trigger safety hazards, extremely high demands are placed on production safety monitoring and early warning systems. Therefore, there is an urgent need for an early warning method with higher accuracy, real-time performance, and intelligence to improve risk prevention and control effectiveness.

[0003] Current mainstream mine safety early warning methods mostly rely on single sensors or limited-dimensional data for threshold determination. They collect environmental data at a fixed frequency, perform simple filtering, and directly compare it to a preset threshold; exceeding the threshold triggers an alarm. These methods fail to fully utilize big data analytics for correlation analysis of multi-source data, making it difficult to identify the coupling relationships and potential risk patterns between environmental parameters. Insufficient data cleaning and inadequate completion of missing values ​​mean risk identification depends on single-point anomalies, failing to assess multiple overlapping risks. Furthermore, the fixed sampling pattern results in insufficient data resolution in high-risk scenarios, leading to delayed warnings and significant false alarms and missed alarms.

[0004] Because existing technologies cannot achieve accurate identification and dynamic real-time early warning of multiple overlapping risks based on multi-source information fusion and big data analysis, existing technologies suffer from insufficient accuracy and real-time performance in early warning. Summary of the Invention

[0005] This invention provides a method and system for early warning of mine production safety, in order to solve the problems of insufficient accuracy and real-time performance of early warning in existing technologies.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for early warning of mine production safety, comprising: Collect raw environmental data from multiple sources, clean the raw environmental data from multiple sources, and obtain a preliminary cleaned dataset; The initial cleaned dataset is smoothed by mean to obtain a continuously smoothed dataset. Missing values ​​in the continuously smoothed dataset are then filled by interpolation to obtain a standard cleaned dataset. Data consistency verification is performed on the standard cleaned dataset to obtain logical correlation results. Based on the logical correlation results, outlier screening is performed on the standard cleaned dataset to obtain multiple risk superposition identifiers. Based on the multiple risk overlay identifiers, a risk identifier value is calculated according to a preset initial weight. If the risk identifier value exceeds a preset risk threshold, the sampling frequency of the corresponding region of the risk identifier value is increased to obtain refined risk data. The volatility characteristics of the detailed risk data are analyzed to obtain volatility detection results. The preset initial weights are updated based on the volatility detection results to obtain a comprehensive risk assessment score. If the comprehensive risk assessment score is higher than the preset warning threshold, a regional warning signal is generated. Based on the regional warning signal, the sensor acquisition deviation is calibrated to generate a real-time warning result. The acquisition density and power consumption allocation are adjusted according to the real-time warning result to determine an optimized multi-source information acquisition mode.

[0007] In a second aspect, the present invention provides a mine production safety early warning system, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any of the above-mentioned embodiments.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the preceding claims.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention uses data cleaning, mean smoothing and missing value interpolation to sequentially filter noise, remove outliers, smooth data and fill gaps in the original environmental data from multiple sources, thereby improving the integrity and continuity of the data and providing a reliable data foundation for subsequent risk identification.

[0010] (2) This invention uses data consistency verification, anomaly screening and multiple risk superposition identification technology, and uses Pearson correlation coefficient to analyze the correlation degree of multi-source data and screen synchronous anomaly feature points. It can accurately identify multiple risk superposition states, improve the accuracy of risk assessment, and reduce the probability of false alarms and false alarms.

[0011] (3) By dynamically increasing the sampling frequency, analyzing fluctuation characteristics and using sensor calibration technology, this invention can adaptively adjust the acquisition strategy according to the risk status and correct the acquisition deviation, thereby improving the real-time performance and data accuracy of early warning, while optimizing the acquisition density and power consumption distribution, and achieving stable and reliable downhole safety early warning. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the mine production safety early warning method provided in the first embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Reference Figure 1 The first embodiment of the present invention provides a method for early warning of mine production safety, comprising the following steps: S11, Collect raw environmental data from multiple sources, perform data cleaning on the raw environmental data from multiple sources, and obtain a preliminary cleaned dataset; S12, the initial cleaned dataset is smoothed by mean to obtain a continuous smoothed dataset, and the missing values ​​of the continuous smoothed dataset are filled by interpolation to obtain a standard cleaned dataset; S13, perform data consistency verification on the standard cleaned dataset to obtain logical correlation results, and perform outlier screening on the standard cleaned dataset based on the logical correlation results to obtain multiple risk superposition identifiers; S14, calculate the risk identifier value based on the multiple risk superimposed identifiers according to the preset initial weight; if the risk identifier value exceeds the preset risk threshold, increase the sampling frequency of the corresponding area of ​​the risk identifier value to obtain refined risk data. S15, perform volatility characteristic analysis on the refined risk data to obtain volatility detection results, update the preset initial weights based on the volatility detection results, and obtain a comprehensive risk assessment score; S16, if the comprehensive risk assessment score is higher than the preset warning threshold, generate a regional warning signal, calibrate the sensor acquisition deviation based on the regional warning signal, and generate a real-time warning result; adjust the acquisition density and power consumption allocation according to the real-time warning result to determine the optimized multi-source information acquisition mode.

[0015] In step S11, it is necessary to collect raw environmental data from multiple sources, and perform data cleaning on the raw environmental data from multiple sources to obtain a preliminary cleaned dataset, including: Raw environmental data from multiple sources is collected through a sensor network deployed underground. This raw environmental data includes gas concentration, dust density, and temperature. The original multi-source environmental data is filtered for noise using a median filter with a fixed window size to obtain a denoised dataset. For the denoised dataset, abnormal fluctuation points are removed by calculating the data distribution interval using the 3σ statistical criterion, resulting in a preliminary cleaned dataset.

[0016] It should be noted that the multi-source environmental raw data refers to data collected by the underground sensor network, which reflects the safety status of the underground environment and includes three types of physical quantities: gas concentration, dust density, and temperature. The sampling frequency of the data is set according to relevant mine safety monitoring specifications, with an initial sampling frequency of once per minute, which can be dynamically adjusted according to subsequent risk conditions. The window size of the median filter is determined based on the common noise fluctuation range of underground data and is fixed at 3. This window length can retain the true trend of data change while filtering out random noise and can be adaptively adjusted according to on-site working conditions. The statistical criteria are based on the statistical law of normal distribution. The specific values ​​are determined by calculating the mean and standard deviation of continuously collected historical normal operating data. The normal data range is the mean ± 3 times the standard deviation. Data outside this range is judged as abnormal fluctuation points. This judgment rule can be adjusted according to different mine geological conditions and equipment characteristics.

[0017] In this step, firstly, according to the underground mining layout and safety monitoring specifications, a sensor network is deployed at key locations prone to safety risks, such as the mining face, return airway, intake airway, electromechanical chamber, ventilation nodes, and transfer points. During deployment, it is ensured that the sensors are securely installed, fixed in position, unobstructed, and free from interference, guaranteeing stable, continuous, and synchronous acquisition of three parameters: gas concentration, dust density, and temperature. After data acquisition begins, the system performs data reading once per minute at the initial sampling frequency, sequentially acquiring the gas concentration, dust density, and temperature values ​​at each monitoring point. Each set of data is automatically stamped with a timestamp, location number, and parameter type identifier, forming structured multi-source environmental raw data.

[0018] Furthermore, after the raw data acquisition is completed, the system performs sequence processing on the multi-source environmental raw data. Median filtering is performed independently on the gas concentration sequence, dust density sequence, and temperature sequence. The filtering window is fixed at 3, that is, taking the current data point as the center, taking one data point forward and one data point backward, sorting the three values ​​in the window according to their size, and taking the median value as the new value after filtering at the current point. The entire sequence is filtered point by point to achieve the removal of random noise, spike interference, and instantaneous change signals, while preserving the overall trend of data change, resulting in a denoised dataset.

[0019] Furthermore, the system retrieves long-term stable normal data for corresponding monitoring points and parameter types from a pre-stored historical normal operating condition database. Based on this data, it calculates the arithmetic mean and standard deviation, and then... Statistical criteria are used to construct normal data intervals: the lower limit of the normal interval is the mean minus three standard deviations, and the upper limit is the mean plus three standard deviations. Each value in the denoised dataset is then compared to its corresponding normal interval. If a value is less than the lower limit or greater than the upper limit, it is identified as an abnormal fluctuation and removed; otherwise, it is retained as valid data. The system then rearranges and combines the remaining valid data according to its original chronological order, restoring the continuity of the data sequence to form a complete, standardized, and reliable preliminary cleaned dataset. This provides a high-quality data foundation for subsequent data smoothing, missing value completion, and association analysis.

[0020] It should be noted that the long-term stable operation mentioned in this invention refers to the period during which environmental parameters in the downhole monitoring area continuously meet the following fluctuation range requirements under the conditions of no abnormal alarms, no production disturbances (such as blasting, support, and transportation operations), and normal equipment operation: gas concentration fluctuation range does not exceed ±5% of the historical average or absolute fluctuation does not exceed ±5ppm (whichever is greater); dust density fluctuation range does not exceed ±10% of the historical average or absolute fluctuation does not exceed ±0.5mg / m³. 3 (The larger value shall be used), and the temperature fluctuation range shall not exceed ±2℃. This upper limit of the fluctuation range is determined comprehensively based on relevant mine safety monitoring standards (such as the provisions on permissible fluctuations in gas concentration under normal ventilation conditions in the "Coal Mine Safety Regulations") and the statistical distribution of historical normal operating data of this mine (taking the 95% confidence interval under normal operating conditions). The minimum duration of a stable state is set to 24 consecutive hours. That is, only when all parameters do not exceed the above fluctuation range for 24 consecutive hours, and there are no records of manual intervention or equipment failure during this period, can this data segment be included in the "Pre-stored Historical Normal Operating Condition Database". The criteria for excluding unstable events include: the existence of alarm records, records of blasting or large equipment start-up and shutdown, sensor disconnection or restart, and manually marked periods as abnormal. The system automatically scans historical data daily, adding data that meets the above stability conditions to the database and removing old data that exceeds the 30-day validity period.

[0021] It is worth noting that the pre-stored historical normal operating condition database adopts a time-series structured architecture. The data is in the form of a standardized sequence with timestamps, location numbers, parameter types, and values. The data source is measured data from the same monitoring point that has been operating stably for a long time. The update mechanism is automatic daily updates, which are used to provide basic data for the calculation of the 3σ criterion. For example, for the gas concentration sensor in area 3 of the well, the historical normal data records stored in the database include: timestamp "2025-05-01 08:00:00", point number "P-03", parameter type "gas concentration", value "50.2"; timestamp "2025-05-01 08:01:00", point number "P-03", parameter type "gas concentration", value "50.5"; timestamp "2025-05-01 08:02:00", point number "P-03", parameter type "gas concentration", value "49.9"; timestamp "2025-05-01 08:03:00", point number "P-03", parameter type "gas concentration", value "50.3"; timestamp "2025-05-01 "08:04:00", location number "P-03", parameter type "gas concentration", value "50.1". Based on this historical data, the system can calculate the mean gas concentration at this location as 50.2, with a standard deviation of 0.22, thus constructing a normal data range of [49.54, 50.86] (mean ± 3 times standard deviation), which is used to determine subsequent abnormal fluctuation points.

[0022] For example, in a downhole environment, a sensor network is deployed to collect multi-source information within a specific area. First, data on gas concentration, dust density, and temperature changes are collected in real time, once per minute. Gas concentration is expressed in ppm, and dust density in mg / m³. 3The units are as follows, with temperature in °C. The collected raw dataset includes a gas concentration sequence of [50.2, 52.1, 49.8, 120.5, 51.3], a dust density sequence of [2.5, 2.7, 2.4, 10.8, 2.6], and a temperature sequence of [25.1, 24.9, 25.3, 35.7, 25.0]. Subsequently, the raw data underwent noise filtering using median filtering with a window size of 3. For each data point, the median of the three points before and after it was taken as the new value. Taking gas concentration as an example, the original sequence's first point retains its original value of 50.2; the second point window [50.2, 52.1, 49.8] has a median value of 50.2 after sorting; the third point window [52.1, 49.8, 120.5] has a median value of 52.1 after sorting; the fourth point window [49.8, 120.5, 51.3] has a median value of 51.3 after sorting; and the fifth point window [120.5, 51.3] has its effective median value of 51.3 after edge processing. After processing in this way, a smooth sequence is obtained: gas concentration [50.2, 50.2, 52.1, 51.3, 51.3], dust density [2.5, 2.5, 2.7, 2.6, 2.6], and temperature [25.1, 25.0, 25.1, 25.3, 25.0]. It can be seen that the impulse noise in the original data, such as 120.5, 10.8, and 35.7, has been effectively filtered out by median filtering.

[0023] Furthermore, for the smoothed dataset after median filtering, a statistical distribution-based approach is adopted. The criteria include outlier detection to verify data consistency. Taking the smoothed gas concentration sequence [50.2, 50.2, 52.1, 51.3, 51.3] as an example, the mean is calculated as (50.2 + 50.2 + 52.1 + 51.3 + 51.3) / 5 = 51.02, and the standard deviation is 0.73 (variance calculation: sum of squares of deviations = 0.6724 + 0.6724 + 1.1664 + 0.0784 + 0.0784 = 2.668, variance = 0.5336, standard deviation = 0.73). The normal range is the mean ± 3 times the standard deviation, i.e., [48.83, 53.21]. All values ​​in the sequence fall within this range, with no abnormal fluctuations, therefore no removal or filling is necessary. The dust density smoothing sequence [2.5, 2.5, 2.7, 2.6, 2.6] was calculated in the same manner: mean 2.58, standard deviation 0.08. The interval [2.34, 2.82] shows all values ​​are normal; for the temperature smoothing sequence [25.1, 25.0, 25.1, 25.3, 25.0], the mean is 25.1 and the standard deviation is 0.11. The range [24.77, 25.43] contains all normal values. The final preliminarily cleaned dataset includes the following sequences: gas concentration [50.2, 50.2, 52.1, 51.3, 51.3], dust density [2.5, 2.5, 2.7, 2.6, 2.6], and temperature [25.1, 25.0, 25.1, 25.3, 25.0]. This processing ensures logical consistency between noise filtering and anomaly detection, providing a reliable foundation for subsequent data smoothing and missing value completion.

[0024] In step S12, the initial cleaned dataset needs to be smoothed by mean to obtain a continuously smoothed dataset, and the continuously smoothed dataset needs to be filled with missing values ​​by interpolation to obtain a standard cleaned dataset, including: The initial cleaned dataset is averaged point by point according to a preset time window to complete the data smoothing process and obtain a continuous smoothed dataset. Traverse the continuous smooth dataset to identify and locate missing data points, and obtain the missing point information; The missing point information is processed by linear interpolation to obtain a standard cleaned dataset.

[0025] It should be noted that the preset time window is determined based on the rate of change of underground environmental parameters. In this embodiment, a window length of 5 minutes is used. This value is derived from the time-domain variation characteristics of underground environmental data. The window length is required to effectively eliminate high-frequency noise without masking the true risk trend, and can be adjusted according to the actual working conditions of the mine. Missing point information refers to the numerical gaps caused by sensor transmission interruptions, equipment failures, etc., including timestamps, monitoring points, and parameter types. The filler values ​​of linear interpolation are calculated from the adjacent valid data before and after the missing point according to a linear relationship, which can restore the true data trend. The pre-stored historical normal working condition database adopts time-series structured storage, and the data source is stable operating data in the same area. The update mechanism is daily automatic updates. This database is used to verify the reasonableness of linear interpolation imputation results. The specific verification method is as follows: the imputed value is compared with the historical normal operating data of the same parameter at that point, and the absolute value of the deviation between the imputed value and the historical mean is calculated. If the deviation does not exceed three times the historical standard deviation (i.e., it is within the 3σ interval of the historical normal data), the imputation result is deemed reasonable and retained. If the deviation exceeds three times the historical standard deviation, it is deemed unreasonable imputation. At this time, the system automatically switches to median interpolation (taking the median of the two valid data before and after the missing point) to recalculate the imputed value and performs 3σ verification again. If the second imputation is still unreasonable, the missing point is marked as "data unreliable" and removed in subsequent risk calculations, and a sensor status check alarm is triggered.

[0026] The pre-stored historical normal operating condition database uses time-series structured storage, with data sourced from stable operating data in the same region. The update mechanism is daily automatic updates (consistent with the aforementioned). This database is used to verify the rationality of linear interpolation filling results. The specific verification method involves comparing the filled value with historical normal operating condition data of the same parameters at that point, calculating the absolute value of the deviation between the filled value and the historical mean. If the deviation does not exceed three times the historical standard deviation (i.e., within the 3σ interval of historical normal data), the filling result is deemed reasonable and retained. If the deviation exceeds three times the historical standard deviation, it is deemed unreasonable filling. In this case, the system automatically switches to median interpolation (taking the median of the two valid data points before and after the missing point) to recalculate the filled value and performs 3σ verification again. If the second filling is still unreasonable, the missing point is marked as "data unreliable," removed from subsequent risk calculations, and a sensor status check alarm is triggered. The three-times-standard-deviation threshold is derived from the statistical principle of covering 99.7% of normal data under a normal distribution and can be adjusted to 2.5 times or 3.5 times according to the actual data distribution characteristics of the mine.

[0027] In this step, the initial cleaned dataset is first subjected to point-by-point mean smoothing calculations within a preset 5-minute time window. The smoothing process centers on the current data point, extracts all valid data points spanning the 5-minute time frame, extracts all values ​​within this window, sums them, and then divides them by the number of valid data points within the window to obtain the arithmetic mean. This arithmetic mean replaces the original data point values, completing the full-sequence smoothing process point by point. This effectively eliminates high-frequency small fluctuations, reduces the impact of residual noise, and improves the continuity and stability of the data curve, resulting in a continuously smoothed dataset.

[0028] Furthermore, after smoothing, the system iterates through the continuously smoothed dataset point by point in timestamp order, checking for empty values, invalid identifiers, and obvious breakpoints. It precisely locates positions with missing values, recording the timestamp, monitoring point number, and parameter type of each missing point to form a complete list of missing point information. For each missing point, the system reads the two adjacent valid values ​​before and after it, constructs a linear relationship using these two values ​​as endpoints, calculates the theoretical filling value for the intermediate missing point according to linear interpolation rules, and writes the calculated filling value into the missing position, ensuring the data sequence is continuous and uninterrupted in the time dimension. After filling is complete, the system performs integrity verification on the entire sequence to confirm no omissions, duplications, or abnormal filling values. This ultimately forms a standardized cleaned dataset with uniform format, continuous values, low noise, and high reliability, used for subsequent multi-parameter consistency verification and identification of multiple risk superpositions.

[0029] For example, for the initially cleaned dataset, further optimization is performed using data smoothing techniques and missing value imputation logic to improve data quality. First, for data smoothing, mean smoothing within a time window is used, with the time window size set to 5 minutes. The gas concentration sequence [50.2, 50.2, 52.1, 51.3, 51.3] is mean-wise calculated to obtain the smoothed sequence [50.2, 50.76, 51.02, 51.28, 51.3]; the dust density sequence [2.5, 2.5, 2.7, 2.6, 2.6] is smoothed to obtain the sequence [2.5, 2.54, 2.58, 2.62, 2.6]; and the temperature sequence [25.1, 25.0, 25.1, 25.3, 25.0] is smoothed to obtain the sequence [25.1, 25.06, 25.1, 25.14, 25.1]. Secondly, linear interpolation is used to fill in any missing values ​​that may exist in the dataset. In the gas concentration sequence, assume there is a missing point between the smoothed value 51.02 (corresponding to time t1) and the value 51.28 (corresponding to time t2). The time corresponding to this missing point is t, and it is located between t1 and t2. When the missing point is exactly in the middle of t1 and t2, (t-t1) / (t2-t1)=0.5, and substituting this into the equation, we get the filled value = 51.02 + (51.28 - 51.02) × 0.5 = 51.15. If the missing point is located at other time positions, it needs to be calculated according to the actual time ratio. For example, if the missing point is located at the end of 1 / 3 of t1, the filled value = 51.02 + (51.28 - 51.02) × (1 / 3) ≈ 51.11. In this example, it is assumed that the missing point is located at the midpoint of time. Therefore, the completed gas concentration sequence is [50.2, 50.76, 51.02, 51.15, 51.28, 51.3], ensuring data continuity and integrity. The same linear interpolation rules are applied to missing values ​​of other parameters.

[0030] In step S13, data consistency verification needs to be performed on the standard cleaned dataset to obtain logical correlation results. Based on the logical correlation results, outlier screening is performed on the standard cleaned dataset to obtain multiple risk superposition indicators, including: For the standard cleaning dataset, the correlation between the gas concentration and the dust density was analyzed using the Pearson correlation coefficient to obtain logistic correlation results; If the logical correlation result meets the preset correlation threshold, the feature points with synchronous abnormal increase are selected from the standard cleaned dataset according to the preset fluctuation threshold to obtain the abnormal synchronous feature points; For the abnormal synchronization feature points, risk levels are marked by combining preset business rules to obtain multiple risk superposition identifiers.

[0031] It should be noted that the Pearson correlation coefficient is used to quantify the linear correlation between gas concentration and dust density, and is established based on the fact that these two parameters often show synchronous changes before the occurrence of downhole risks. The preset correlation threshold is determined based on statistical analysis of normal operating data, specifically the average Pearson correlation coefficient between gas concentration and dust density in the normal operating data. The normal operating data mentioned here adopts the quantitative definition of long-term stable operation mentioned above in this invention, namely, no abnormal alarms, no production disturbances, normal equipment operation, gas concentration fluctuation ≤ historical average ±5% or ±5ppm, and dust density fluctuation ≤ historical average ±10% or ±0.5mg / m³. 3 Temperature fluctuation ≤ ±2℃, and continuous stable duration ≥ 24 hours.

[0032] To ensure the statistical stability of the mean correlation coefficient, the sample used to calculate this mean should meet the following requirements: continuous collection for no less than 30 calendar days (a time span of at least one month), and the number of valid data points per day should be no less than 95% of the theoretical number of sampling points for that day (for example, if the initial sampling frequency is 1 time / minute, then the theoretical number of sampling points per day is 1440, and the actual number of valid points should be ≥1368). The mean correlation coefficient is automatically recalculated every 30 days, and the preset correlation threshold is updated accordingly to adapt to the long-term slow changes in the underground environment. In this embodiment, the calculated mean correlation coefficient is 0.8, so the preset correlation threshold is set to 0.8. This value can be adjusted according to the geological conditions and equipment characteristics of different mines.

[0033] It should be noted that the preset fluctuation threshold is determined by adding 1.5 times the standard deviation to the mean of the historical normal data of each parameter, and is used to identify abnormal increases. It can be adjusted according to the equipment accuracy. The preset business rules are based on relevant national mine safety monitoring standards and the actual risk evolution patterns underground, classifying abnormal points into three risk levels: low, medium, and high.

[0034] It should be noted that different standard deviation multiples are used in different stages of this invention, with different technical purposes. In step S11, 3 times the standard deviation (mean ± 3σ) is used as the anomaly rejection boundary to remove extreme outliers caused by accidental sensor malfunctions, instantaneous shocks, etc. These extreme values ​​have a probability of less than 0.3% under the normal distribution assumption, and their removal ensures the robustness of subsequent data statistics. In this step (S13), the mean + 1.5 times the standard deviation is used as the anomaly rise threshold to detect the synchronous rise trend of gas concentration and dust density that has not yet reached extreme levels but has significantly deviated from the normal range. This threshold should be more sensitive than the rejection threshold so as to capture abnormal synchronous feature points in the early stage of risk development.

[0035] Based on data from 47 confirmed abnormal events in this mine over the past three years (covering single and double anomalies in gas concentration or dust density), ROC curve analysis was used to calculate the true positive rate (sensitivity) and false positive rate (1-specificity) for the detection target of "synchronous abnormal increase within 30 minutes before the event" when using different standard deviation multiples (1.0σ, 1.2σ, ​​1.5σ, 1.8σ, 2.0σ, 2.5σ, 3.0σ) as anomaly thresholds. Maximizing the Youden index (sensitivity + specificity - 1) determined the optimal multiple to be 1.5σ, at which point the sensitivity was 92% and the false positive rate was 8%. Simultaneously, the same multiple test was performed on historical stable operating data (24 consecutive hours without alarms or disturbances), and the false alarm rate was less than 5%. Considering both the timeliness of early warning and the risk of false alarms, this embodiment selected 1.5 times the standard deviation as the abnormal increase threshold.

[0036] The specific classification rules are as follows: For gas concentration, the historical average is used as the benchmark. A value exceeding 1.5 times but less than 2 times the average is considered slightly abnormal; 2 to 3 times is considered moderately abnormal; and more than 3 times is considered highly abnormal. For dust density, the same 1.5, 2, and 3 times standard deviations are used to classify it as slightly, moderate, or highly abnormal. The comprehensive risk level of multiple risk indicators is determined by the combination of the degree of abnormality of gas concentration and dust density: if both are slightly abnormal and the duration does not exceed 2 minutes, it is considered low risk; if at least one is moderately abnormal, or both are slightly abnormal and the duration exceeds 5 minutes, it is considered medium risk; if at least one is highly abnormal, or both are moderately abnormal and the duration exceeds 3 minutes, or a combination of moderate gas concentration abnormality and slight dust density abnormality and the duration exceeds 5 minutes, it is considered high risk.

[0037] The duration thresholds (2 minutes, 3 minutes, and 5 minutes) mentioned in the above-mentioned grading rules were determined based on statistical analysis of 47 confirmed abnormal events related to gas concentration and dust density in this mine over the past three years (12 low-risk events, 21 medium-risk events, and 14 high-risk events). The statistical method was to extract the moment when the gas concentration and dust density first simultaneously exceeded their respective abnormal thresholds (1.5 times the standard deviation) before each event occurred as the starting point, and the moment when both returned to the normal threshold range (below 1.5 times the standard deviation) as the ending point, and then calculate the duration of the synchronous abnormality. A percentile analysis was performed on the duration distribution of low-risk events (those that do not develop into accidents and only require recording). The 90th percentile was set at 2 minutes, meaning that 90% of low-risk events last no more than 2 minutes. For medium-risk events (requiring manual intervention or equipment maintenance), the 80th percentile was set at 5 minutes. For high-risk events (directly triggering alarms or causing accidents), the 70th percentile was set at 3 minutes. Considering both safety redundancy (appropriately shortening the high-risk threshold for early warning) and false alarm prevention requirements (extending the low-risk threshold to avoid frequent alarms), the final thresholds were determined to be 2 minutes for low-risk events, 5 minutes for medium-risk events, and 3 minutes for high-risk events.

[0038] The counting start point for the duration of anomalies is the time when both parameters are first detected to simultaneously exceed their respective anomaly thresholds, and the end point is the time when both parameters return to the normal threshold range. For example, if the gas concentration at a monitoring point exceeds the mean by 2.2 standard deviations (moderate) and the dust density exceeds the mean by 1.6 standard deviations (mild), and the duration of both anomalies is 6 minutes, it is judged as high risk according to the above rules. The multiple risk overlay indicator is used to characterize the coupled risk state of simultaneous anomalies in multiple parameters, including information such as location number, anomaly start time, anomaly duration, gas concentration anomaly level, dust density anomaly level, and overall risk level.

[0039] In this step, gas concentration and dust density sequences are first extracted from the standard cleaning dataset. The two sequences are then aligned by timestamps to ensure comparative analysis across the same time dimension. Subsequently, the covariance and standard deviation of the two sequences are calculated using the Pearson correlation coefficient. A linear correlation coefficient is obtained through standardization, with values ​​ranging from -1 to 1. A value closer to 1 indicates a stronger positive correlation between the two parameters. The system compares the calculated correlation coefficient with a preset correlation threshold of 0.8. If the correlation coefficient is greater than or equal to 0.8, the two parameters are considered to have a significant linear relationship, meeting the prerequisite for multiple risk superposition. If the correlation coefficient is less than 0.8, the correlation is considered weak, and no synchronous outlier screening is performed.

[0040] Furthermore, after the correlation verification passes, the system reads the historical normal data for gas concentration and dust density, calculates their respective mean and standard deviation, and obtains the abnormal fluctuation threshold for the corresponding parameter by adding 1.5 times the standard deviation to the mean. Then, the two sets of data are compared point by point. If the gas concentration value and dust density value both exceed their respective abnormal thresholds at a certain moment and show a synchronous upward trend, then the data point at that moment is determined as an abnormal synchronous feature point.

[0041] Furthermore, the system performs duration statistics and magnitude calculations on the identified abnormal synchronization feature points. Combined with the classification rules for low, medium, and high risks in the national mine safety monitoring standards, the system labels the abnormal synchronization feature points with risk levels, forming a multi-layered risk label that includes location, time, anomaly type, risk level, and parameter combination. This provides a clear basis for subsequent risk quantification calculations and sampling frequency control.

[0042] For example, based on the standard cleaning dataset obtained in step S12, namely the gas concentration sequence [50.2, 50.76, 52.5, 51.28, 51.3], the dust density sequence [2.5, 2.54, 2.9, 2.62, 2.6], and the temperature sequence [25.1, 25.06, 25.2, 25.14, 25.1], where the third data point is set according to the actual monitored abnormal increase, and the remaining points are inherited from the smoothed output of step S12, data consistency verification is performed. The Pearson correlation coefficient is used to analyze the correlation between gas concentration and dust density, and the calculated correlation coefficient is 0.94, which is higher than the preset correlation threshold of 0.8, thus the verification passes. The calculated mean gas concentration is (50.2+50.76+52.5+51.28+51.3) / 5=51.208, with a standard deviation of 0.86; the mean dust density is (2.5+2.54+2.9+2.62+2.6) / 5=2.632, with a standard deviation of 0.15. The abnormal rise threshold is set to the mean + 1.5 times the standard deviation. The gas concentration threshold is 51.208+1.29=52.498, and the dust density threshold is 2.632+0.225=2.857. In the data sequence, the third data point has a gas concentration of 52.5 (exceeding 52.498) and a dust density of 2.9 (exceeding 2.857), both simultaneously exceeding the abnormal threshold.

[0043] Furthermore, according to preset business rules, the gas concentration exceeded the mean (52.5-51.208) / 0.86≈1.5 times the standard deviation, which is considered a slight anomaly; the dust density exceeded (2.9-2.632) / 0.15≈1.79 times the standard deviation, which is also considered a slight anomaly. Continuous monitoring found that the anomaly lasted for 3 minutes (all three sampling points were in an abnormal state). Based on the classification rules (both are slight anomalies and the duration exceeded 2 minutes but did not reach 5 minutes), it was comprehensively judged as a medium risk. The system automatically marked this point as a multiple risk superposition indicator, including the point number "P-03", the anomaly start time "14:35:20", the duration of 180 seconds, the gas anomaly level "slight", the dust anomaly level "slight", and the overall risk level "medium", and pushed it to the subsequent early warning module. Through the above automated process, from correlation analysis to anomaly detection to risk identification, a complete logical chain is formed to ensure the scientificity and reliability of data analysis and provide accurate support for downhole environmental monitoring.

[0044] In step S14, a risk identifier value needs to be calculated based on the multiple risk overlay identifiers according to a preset initial weight. If the risk identifier value exceeds a preset risk threshold, the sampling frequency of the corresponding region of the risk identifier value is increased to obtain refined risk data, including: Based on the aforementioned multiple risk superposition indicators, the risk indicator value is obtained by weighting the gas concentration, the dust density, and the temperature according to a preset initial weight. The risk identification value is compared with a preset risk threshold. If the risk identification value exceeds the preset risk threshold, the deviation ratio between the risk identification value and the average of the pre-stored historical data is calculated. Based on the deviation ratio, the sensor sampling frequency in the area corresponding to the risk identification value is dynamically increased to generate a high-frequency acquisition strategy; Data is collected according to the high-frequency acquisition strategy to obtain detailed risk data with high temporal resolution.

[0045] It should be noted that the initial weights are determined based on the degree of influence of the three types of parameters on mine safety, using the Analytic Hierarchy Process (AHP) combined with expert scoring and historical accident causation statistics. First, a hierarchical structure is constructed between the target layer (mine production safety risk) and the criterion layer (gas concentration, dust density, and temperature). At least five mine safety experts are invited to conduct pairwise comparisons of the three types of parameters using the Saaty 1-9 scaling method. For example, compared to dust density, gas concentration is considered slightly more important and assigned a value of 2; compared to temperature, gas concentration is considered significantly more important and assigned a value of 5; compared to temperature, dust density is considered slightly more important and assigned a value of 2. This results in the construction of a judgment matrix A = [[1,2,5],[1 / 2,1,2],[1 / 5,1 / 2,1]]. The largest eigenvalue and corresponding eigenvector of this matrix are calculated, and after normalization, the initial AHP weight vector is obtained as (0.581,0.309,0.110). The consistency test showed a consistency ratio of CR = 0.003 < 0.1, which is considered satisfactory.

[0046] It should be noted that the initial weights were determined using the Analytic Hierarchy Process (AHP) based on engineering experience in actual underground management where gas concentration and dust density are given equal importance, and temperature has a relatively minor impact. A hierarchical structure was constructed between the target layer (mine production safety risk) and the criteria layer (gas concentration, dust density, and temperature). At least five mine safety experts were invited to conduct pairwise comparisons using the Saaty 1-9 scaling method: gas concentration and dust density were equally important, assigned a value of 1; gas concentration was significantly more important than temperature, assigned a value of 5; and dust density was significantly more important than temperature, assigned a value of 5. The resulting judgment matrix B is as follows:

[0047] Calculate the largest eigenvalue of the matrix. The corresponding feature vector, after normalization, yields the weight vector as follows: Consistency Indicators Random Consistency Index Consistency ratio The consistency check was passed. Therefore, the preset initial weights were determined to be 0.4 for gas concentration, 0.4 for dust density, and 0.2 for temperature. The following embodiments will use this set of preset initial weights as examples. These weights can be recalculated and adjusted using the same AHP process according to the disaster type of different mines. For example, the gas concentration weight can be appropriately increased for gas outburst mines, and the dust density weight can be appropriately increased for dust explosion risk mines.

[0048] The high-frequency acquisition strategy increases the sampling frequency based on the deviation ratio, making the frequency positively correlated with the risk level. The new sampling frequency is equal to the sum of the base frequency multiplied by 1 and the deviation rate divided by 100. This calculation rule is based on the positive linear correlation between the underground risk level and the parameter fluctuation amplitude. It can realize the synchronous linear increase of the sampling frequency with the risk deviation level, which can not only ensure the time resolution and monitoring sensitivity of data acquisition in high-risk scenarios, but also avoid the waste of system power consumption caused by meaningless over-frequency acquisition. At the same time, it conforms to the general design logic of dynamic sampling control of mine safety monitoring system.

[0049] In this step, the monitoring points corresponding to the multiple risk overlay indicators are first read, and the effective values ​​of gas concentration, dust density, and temperature at the same time are extracted from these points. A weighted summation operation is then performed according to preset initial weights, where gas concentration has a weight of 0.4, dust density has a weight of 0.4, and temperature has a weight of 0.2. The weighted sum of these three values ​​yields the risk indicator value. After calculation, the system compares the risk indicator value with a preset risk threshold of 0.8. If the risk indicator value is less than or equal to 0.8, the current risk level is determined to be normal, and the original sampling frequency remains unchanged. If the risk indicator value is greater than 0.8, the current area is determined to be in a high-risk state.

[0050] It should be noted that, based on a retrospective analysis of 15 typical high-risk accident cases in this mine over the past three years, the risk indicator values ​​(calculated according to the aforementioned weights) within 15 minutes prior to the accident were extracted. The minimum value was 0.78 before a gas outburst accident, the maximum value was 0.92 before a dust explosion accident, and the average value was approximately 0.85. Taking into account both safety margin and false alarm prevention requirements, the threshold was set at 0.8. That is, when the risk indicator value exceeds 0.8, the current area is determined to be in a high-risk state.

[0051] Furthermore, the degree of risk deviation is calculated by subtracting the average of pre-stored historical data from the current risk indicator value, then dividing the difference by the historical data average to convert it into a percentage deviation ratio. The system adjusts the sampling frequency according to the deviation ratio using a preset linear rule. The new frequency is calculated by multiplying the base frequency by 1 and dividing the deviation rate by 100, achieving adaptive adjustment where higher risks result in denser sampling. Adjustment commands are sent to the corresponding sensor nodes, which then perform data acquisition at the new frequency, shortening the sampling interval, improving temporal resolution, and acquiring denser, more refined risk data that better reflects rapid short-term changes. This provides high-density data support for subsequent fluctuation characteristic analysis and comprehensive risk assessment.

[0052] In step S15, volatility characteristic analysis needs to be performed on the refined risk data to obtain volatility detection results. Based on these results, the preset initial weights are updated to obtain a comprehensive risk assessment score, including: For the aforementioned detailed risk data, time series analysis is used to calculate short-term volatility characteristics and obtain volatility characteristic parameters; The volatility characteristic parameters are compared with a preset volatility threshold to obtain the volatility detection result; The preset initial weights are dynamically adjusted based on the volatility detection results to obtain optimized weights; Unify the sampling frequency, physical unit, and timestamp format of the detailed risk data to achieve format standardization and obtain standard format risk data; Based on the standard format risk data, the gas concentration, dust density, and temperature are weighted according to the optimized weights to obtain a comprehensive risk assessment score.

[0053] It should be noted that the preset fluctuation threshold is determined through statistical fitting of long-term stable, risk-free operating data at the same location. The specific determination process is as follows: Data from a period of 30 consecutive days (24 hours per day) at the monitoring point without any abnormal alarms, production disturbances, or abnormal equipment operation is selected as the normal operating condition sample. The sampling frequency is once per minute, ensuring a sample size of no less than 43,200 (30 days × 24 hours × 60 minutes). The standard deviations of the three parameters—gas concentration, dust density, and temperature—are calculated separately. Based on the typical fluctuation characteristics of downhole environmental parameters and engineering experience, the upper limit of normal fluctuation is set at 3 times the standard deviation. The selection of this multiple, k=3, is based on the assumption of a normal distribution. The range covers approximately 99.7% of normal data; fluctuations outside this range are highly likely to be abnormal. This statistical threshold is widely used in the field of mine safety monitoring. A practical calculation example based on normal operating data from a typical mine: For gas concentration, the standard deviation is 0.1, so the threshold is 0.3; for dust density, the standard deviation is approximately 0.067, so the threshold is 0.2; for temperature, the standard deviation is 0.133, so the threshold is 0.4. Different mines should recalculate the standard deviations based on their own normal operating data (more than 30 days) and use the same k=3 principle to determine the fluctuation threshold for their specific mine. The values ​​in this example should not be directly applied.

[0054] Regarding the adjustment step size for the optimized weights, this embodiment adopts a fixed step size of 0.1. First, through sensitivity analysis experiments, under typical risk evolution scenarios (e.g., the process of gas concentration gradually increasing from normal values ​​to exceeding the threshold by two standard deviations), the impact of step sizes of 0.05, 0.1, and 0.2 on the convergence speed of the comprehensive risk assessment score was tested. Experimental results show that with a step size of 0.05, it takes 5 assessment cycles (30 seconds per cycle) to adjust the weight from 0.4 to 0.65, with a response lag of approximately 150 seconds; with a step size of 0.2, the weight jumps from 0.4 to 0.8 within 2 cycles, causing violent oscillations in the assessment score (fluctuations exceeding 15% between adjacent cycles), easily leading to false alarms; with a step size of 0.1, the target weight range can be reached in 3 cycles, and the fluctuation in the assessment score is controlled within 5%, balancing response speed and stability. Secondly, a step size of 0.1 meets the general requirements for parameter adjustment resolution in mine safety monitoring systems (the minimum adjustment unit for most industrial controllers is 0.1), facilitating engineering implementation. Finally, this step size can be configured within the range of 0.05 to 0.2 according to the actual needs of the mine. In this embodiment, 0.1 is taken as the default recommended value. The following examples all use this step size for dynamic weight adjustment.

[0055] In this step, for detailed risk data, the moving standard deviation is calculated for the three sets of sequences: gas concentration, dust density, and temperature. The degree of data dispersion within a short time window reflects the severity of fluctuations, thus obtaining fluctuation characteristic parameters. Each fluctuation characteristic parameter is compared one by one with a corresponding preset fluctuation threshold: 0.3 for gas concentration, 0.2 for dust density, and 0.4 for temperature. If the fluctuation characteristic parameter is greater than the corresponding threshold, it is determined that the parameter has abnormal fluctuations; if it is less than or equal to the corresponding threshold, it is determined to be in a normal stable state.

[0056] Furthermore, based on the volatility detection results, the system dynamically adjusts the weights, increasing the weights of parameters with abnormal volatility by 0.1 from the initial weights and simultaneously decreasing the weights of stable parameters by 0.1. After the adjustment, the sum of the weights of all parameters remains 1, forming optimized weights that better reflect the current risk status. Even further, the system standardizes the format of the detailed risk data, unifying the sampling frequency to 30 seconds per sample and the physical units to ppm and mg / m³. 3 The system standardizes the timestamp format (℃) to eliminate format differences between data from different sources, resulting in standard-format risk data. Subsequently, the system performs a weighted summation calculation on the standard-format risk data according to optimized weights, yielding a comprehensive risk assessment score. A higher score indicates a higher security risk in the current area. This score will be directly used for subsequent early warning judgments and tiered response, achieving adaptive optimization and precise quantification of the risk assessment model.

[0057] In step S16, if the comprehensive risk assessment score is higher than the preset warning threshold, a regional warning signal is generated. The sensor acquisition deviation is calibrated based on the regional warning signal to generate a real-time warning result. The acquisition density and power consumption allocation are adjusted according to the real-time warning result to determine the optimized multi-source information acquisition mode.

[0058] Wherein, if the comprehensive risk assessment score is higher than a preset warning threshold, a regional early warning signal is generated, and the sensor acquisition deviation is calibrated based on the regional early warning signal to generate a real-time early warning result, including: If the comprehensive risk assessment score is higher than the preset warning threshold, a high-priority early warning signal for the corresponding downhole area is generated to obtain the regional early warning signal. Based on the aforementioned regional early warning signal, the sensor's acquisition deviation is corrected through adaptive sensitivity calibration and environmental compensation to obtain calibrated measured data. The calibration data is compared and analyzed with pre-stored historical anomaly events to obtain real-time early warning results.

[0059] It should be noted that adaptive sensitivity calibration adjusts the sensitivity based on the sensor's historical calibration data and environmental interference to eliminate drift errors. This embodiment employs a drift correction method based on ambient temperature compensation. Under risk-free operating conditions, the sensor automatically performs zero-point calibration every 24 hours, recording the zero-point drift. During actual measurement, the system reads the current ambient temperature and compares it with the temperature at the time of the last calibration. According to the sensor's factory calibration data, for every 1 degree Celsius change in temperature, the measured value will produce a relative error of 0.2% (temperature drift coefficient). Therefore, the system linearly corrects the original measured value based on the difference between the current temperature and the calibration temperature: if the current temperature is higher than the calibration temperature, the measured value is proportionally lowered; if the current temperature is lower than the calibration temperature, the measured value is proportionally increased. Based on the temperature correction, the zero-point drift recorded in the most recent zero-point calibration is subtracted to obtain the calibrated measured data.

[0060] The calibration trigger conditions are categorized into three types: first, periodic triggering, which automatically performs zero-point calibration and temperature coefficient verification every 24 hours; second, event-driven triggering, which immediately initiates temporary calibration when the ambient temperature changes by more than ±5 degrees Celsius or the humidity changes by more than ±20%; and third, after a warning signal is generated, the system forcibly triggers a sensor calibration to ensure data accuracy under high-risk conditions. The calibration coefficient update employs a smoothing filter method. After each calibration, the calculated drift and temperature correction coefficient are weighted and averaged with historical values, with historical values ​​accounting for 80% and the current value accounting for 20%. This smoothing process avoids drastic jumps in calibration coefficients caused by single-measurement noise or instantaneous interference, while gradually tracking the long-term drift trend of the sensor. Different sensor models have different temperature drift coefficients due to variations in manufacturing processes and materials. These coefficients must be re-measured through laboratory calibration experiments before installation; the 0.2% value in this embodiment cannot be directly applied. The calibration experiment should be conducted in a constant temperature chamber with different temperature points, recording the deviation between the sensor output and the standard value. The temperature drift coefficient specific to this sensor is obtained through linear fitting.

[0061] Environmental compensation is based on the collected real-time humidity and air pressure correction measurements. For gas concentration measurements, the humidity compensation correction formula is as follows:

[0062] in, This represents the humidity-compensated gas concentration value (ppm). This is the original sensor measurement (ppm). The current ambient relative humidity (%RH) The reference humidity (%RH) used in the laboratory calibration is 40%RH in this example. The constant 0.002 is the humidity influence coefficient (unit: 1 / %RH) obtained through the laboratory calibration experiment. Specifically, the calibration experiment conditions are as follows: temperature 25℃±2℃, humidity set to five gradients (20%, 40%, 60%, 80%, and 95%) using a constant temperature and humidity chamber, and methane standard gas concentrations (0.5%, 1.0%, 2.0%, and 4.0%, corresponding to 5000ppm, 10000ppm, and 20000ppm respectively). (m, 40000ppm); Under different humidity conditions, the measured values ​​of the gas concentration sensor were recorded, resulting in five sets of data points: at 20% humidity, the standard value of 5000ppm corresponds to a measured value of 4980ppm; at 40% humidity, the measured value is 4950ppm; at 60% humidity, the measured value is 4900ppm; at 80% humidity, the measured value is 4830ppm; and at 95% humidity, the measured value is 4720ppm. With humidity as the independent variable and the deviation rate between the measured value and the standard value as the dependent variable, linear fitting was used to obtain a deviation rate of 0.002 times the humidity, and the goodness of fit R... 2A coefficient of 0.96 indicates a good linear relationship and effectiveness within the 20%-95% humidity range. Different sensor models require recalibration; this coefficient cannot be directly applied.

[0063] If the influence of air pressure is also considered, an air pressure correction term can be added to the humidity compensation:

[0064] in, The pressure influence coefficient (unit: 1 / kPa) is the parameter value provided by the sensor manufacturer's factory calibration. If the sensor manual does not provide this parameter or the field operating conditions differ significantly from the factory calibration conditions, an empirical value of 0.001 / kPa can be used as the initial setting. This value is based on statistical analysis of the pressure response characteristics of most electrochemical gas sensors, and the additional error introduced in the range of 85 to 115 kPa does not exceed ±1%. During subsequent periodic calibration, the actual measurement deviation will be adjusted accordingly. Fine-tuning is all that's needed; The current ambient air pressure (kPa) The standard atmospheric pressure is 101.325 kPa. In this embodiment, if the air pressure fluctuation exceeds ±5 kPa, the air pressure correction is activated; otherwise, it can be ignored. The system calculates the compensation coefficient according to the above formula based on the real-time collected humidity and air pressure data, corrects the original measured value of gas concentration, eliminates the measurement error caused by environmental factors, and obtains calibrated measured data.

[0065] The pre-stored historical abnormal event database adopts a standardized time-series architecture. Each record contains the risk level (high risk / medium risk / low risk), parameter sequence, event type, and occurrence time. Example event record as follows: Risk level "High Risk", Event type "Anomalous Gas Outburst", Occurrence time "2024-08-15 13:22:30", Gas concentration sequence [45,48,52,58,67,79,88,95,102,108,115,120,118,112,105,96,84,72,61,53] (unit ppm) in parameter sequence, Dust density sequence [2.1,2.2,2.3,2.5,2.6,2.8,3.0,3.2,3.3,3.5,3.6,3.7,3.6,3.4,3.2,3.0,2.8,2.6,2.4,2.2] (unit mg / m³) 3The temperature sequence is [25.0, 25.1, 25.2, 25.3, 25.5, 25.6, 25.8, 26.0, 26.2, 26.5, 26.7, 26.9, 27.0, 27.1, 27.0, 26.8, 26.6, 26.4, 26.2, 26.0] (unit: °C). All event records in this database are derived from officially confirmed accident investigation reports from the mine (including accidents with and without injuries) and major near misses manually marked by the safety management department (referring to events that did not cause actual losses but posed significant risks, confirmed by on-site safety officers, team leaders, or monitoring center personnel based on post-event review results).

[0066] In this step, the comprehensive risk assessment score is first compared with the preset warning threshold. If the comprehensive risk assessment score is lower than or equal to the preset warning threshold, the normal monitoring status is maintained and no warning signal is generated. If the comprehensive risk assessment score is higher than the preset warning threshold, it is immediately determined that there is a significant safety risk in the current area, a high-priority area warning signal is generated, and the location, time, risk level and risk type are marked simultaneously.

[0067] Furthermore, after generating the warning signal, the system initiates the sensor calibration process, reads the current sensor's historical calibration records, drift amount, and environmental interference amount, and performs adaptive sensitivity calibration to correct the zero-point drift and sensitivity offset caused by long-term sensor operation. Simultaneously, the system collects real-time environmental humidity and air pressure data, calculates compensation coefficients according to environmental compensation rules, and corrects the original measurements of gas concentration and dust density to eliminate measurement errors caused by changes in temperature, humidity, and air pressure, thus obtaining the calibrated measured data.

[0068] Furthermore, after calibration, the system inputs the measured calibration data into the matching engine and compares it with the pre-stored historical abnormal events in the pre-stored historical abnormal event database. It calculates the similarity of parameter sequences, fluctuation trends, and risk combinations. Based on the similarity, it judges the risk type and degree of danger. Combining multiple risk superposition states, fluctuation characteristics, and location information, it finally generates real-time early warning results that include risk level, dangerous location, change trend, and handling suggestions, providing accurate, reliable, and actionable early warning information for on-site safety management.

[0069] It should be noted that when the system compares the calibrated measured data with the pre-stored historical anomaly event database, it uses dynamic time warping to calculate the similarity. Dynamic time warping can handle time series of different lengths or with time axis offsets. By finding the optimal alignment path between two series, it calculates the cumulative distance and normalizes the distance to a similarity percentage from 0 to 100%. The system calculates the similarity independently for gas concentration series, dust density series, and temperature series, and then takes a weighted average as the comprehensive similarity. The weights are consistent with the initial weights in the risk assessment (0.4 for gas, 0.4 for dust, and 0.2 for temperature). The mapping relationship between similarity intervals and warning levels is as follows: when the comprehensive similarity is ≥90%, it is judged as a Level 1 warning; when the comprehensive similarity is ≤90%, it is judged as a Level 2 warning; when the comprehensive similarity is ≤75%, it is judged as a Level 3 warning; when the comprehensive similarity is <60%, no warning based on historical events is triggered, but monitoring still needs to be continued in conjunction with other risk judgment logic. In this embodiment, the current combination of environmental parameters has an 82% similarity to historical abnormal events, falling within the level-two warning range, thus generating a level-two warning. This mapping relationship can be locally adjusted based on the historical data accumulation and safety management requirements of different mines.

[0070] The three thresholds (90%, 75%, and 60%) were determined based on the following: 217 labeled event records from a pre-stored historical abnormal event database were used to calculate the DTW comprehensive similarity between each event and other events of the same risk level in the database, thus obtaining the similarity distribution within each risk level. Statistical results show that the mean DTW comprehensive similarity within high-risk events is 92%, with a standard deviation of 4%, meaning that over 95% of high-risk events have a similarity of no less than 90%; the mean similarity within medium-risk events is 78%, with a standard deviation of 5%, meaning that over 95% of medium-risk events have a similarity of no less than 75%; and the mean similarity within low-risk events is 65%, with a standard deviation of 6%, meaning that over 95% of low-risk events have a similarity of no less than 60%.

[0071] Meanwhile, cross-similarity analysis between different risk levels shows that the average similarity between high-risk and medium-risk events is 68%, and between medium-risk and low-risk events it is 54%, with no overlap exceeding 70% between levels. Therefore, using 90%, 75%, and 60% as grading thresholds can achieve a matching consistency of over 95% within each level while maintaining good separability between levels. Different mines can redetermine the thresholds using the same method (calculating the 5th percentile or mean minus 1.5 standard deviation of the similarity within each risk level) based on their own accumulated labeled event data, or they can refer to the values ​​in this embodiment as initial values.

[0072] The step of adjusting the acquisition density and power consumption allocation based on the real-time early warning results to determine the optimized multi-source information acquisition mode includes: Based on the risk level of the real-time early warning results, the data collection priorities for different areas are divided to obtain a collection priority sequence; The sensor acquisition density and energy consumption quota are allocated according to the acquisition priority sequence to obtain a resource allocation scheme; Configure the sensor network according to the resource allocation scheme and determine the optimized multi-source information acquisition mode.

[0073] It should be noted that the risk level is divided into multiple levels based on real-time early warning results, with higher levels indicating higher risk. Data collection priority is positively correlated with the risk level, with the area corresponding to the risk indicator value having the highest priority. This classification is based on the management regulations for key mining areas.

[0074] Energy consumption quotas are allocated in a normalized manner based on the total system power consumption and regional priority weights. Specifically, the downhole monitoring area is divided into two categories: high-risk warning areas and regular safety areas. Each category has a corresponding priority allocation coefficient. The system first sums the allocation coefficients for all online areas to obtain a total coefficient value. The actual energy consumption quota for a single area equals the total system power consumption multiplied by the ratio of the area's allocation coefficient to the total coefficient value. This ensures that the sum of energy consumption quotas for all areas strictly equals the total system power consumption, satisfying the normalization mathematical specifications for power consumption allocation. The allocation coefficient is calculated based on the expected energy consumption proportion of each area. Expected energy consumption is mainly determined by the sampling frequency, while also considering data transmission, processing, and standby power consumption. The sampling frequency for high-risk areas is set five times that of regular areas. Under the premise of the same energy consumption per sampling and transmission, the energy consumption ratio per unit time between the two is 5:1. If only these two areas exist, the total energy consumption share is six parts, with the high-risk area accounting for approximately five-sixths and the regular area accounting for approximately one-sixth. In this embodiment, the theoretical allocation coefficient of the high-risk area (based solely on the sampling frequency ratio, with the sampling frequency of the high-risk area being 5 times that of the regular area) is calculated as follows: Assuming the basic energy consumption share of the regular area is 1, then the basic energy consumption share of the high-risk area is 5, the total share is 6, and the theoretical proportion of the high-risk area is 5 / 6≈0.8333. Considering the specific quantification of additional power consumption overhead, AES-128 encryption is applied to each sampled data, resulting in an actual increase in energy consumption of approximately 10%; using a lightweight compression algorithm (such as LZ4) increases energy consumption by approximately 5%; due to the increased sampling frequency, the number of wireless module wake-up times increases synchronously, resulting in an additional increase in energy consumption of approximately 2%; the high underground temperature (exceeding 30°C) causes a decrease in the efficiency of the sensor and wireless module, resulting in an actual increase in power consumption of approximately 5%; the comprehensive additional overhead coefficient is 1.10×1.05×1.02×1.05≈1.238.

[0075] After weighting and adjusting the theoretical proportions, the actual energy consumption share of the high-risk area is 5 × 1.238 = 6.19, and the basic energy consumption share of the normal area is 1 × 1.0 = 1.0 (the normal area does not consider additional overhead such as encryption and compression, but retains standby power consumption and self-test power consumption as a fixed base included in the basic share). The total energy consumption share is 6.19 + 1.0 = 7.19. Therefore, the normalized allocation coefficient is 6.19 / 7.19 ≈ 0.861 for the high-risk area and 1.0 / 7.19 ≈ 0.139 for the normal area, with a sum of 1.

[0076] In this embodiment, for ease of engineering implementation, the allocation coefficient for high-risk areas is rounded to 0.85, and the corresponding allocation coefficient for regular areas is 1-0.85=0.15, with a deviation of approximately 7.9%, which is within the allowable range for engineering (usually <10%). If higher accuracy is required, the theoretically calculated values ​​of 0.861 and 0.139 can be used directly for allocation. In practical applications, the above overhead coefficients should be recalculated based on measured data such as the sensor model, encryption algorithm, and ambient temperature and humidity to determine the allocation coefficients.

[0077] The energy consumption per sampling cycle is determined by the sensor model and transmission protocol. Environmental correction factors can be determined by referring to tables based on parameters such as downhole temperature, humidity, and air pressure; for example, power consumption increases by 5% for every 10 degrees Celsius increase in temperature. The allocation coefficients of 0.85 and 0.15 in this embodiment are merely examples; in actual applications, they should be recalculated based on the on-site sensor configuration and measured power consumption data. Sampling density refers to the number of samples taken per unit time; the density is increased in areas corresponding to risk identification values. The resource allocation scheme follows the principle of prioritizing key areas.

[0078] In this step, the downhole monitoring areas are first divided into high-risk areas (those corresponding to risk indicator values) and regular safe areas (the rest are designated as safe areas) based on the risk level of the real-time early warning results. Higher risk levels correspond to higher acquisition priority, forming an ordered acquisition priority sequence. According to this priority sequence, the system performs energy consumption quota allocation. First, the allocation coefficients of all online areas are summed to obtain a total coefficient. Then, the total power consumption is multiplied by the ratio of the allocation coefficient of a single area to the total coefficient to obtain the actual available energy consumption quota for that area. This ensures that the total energy consumption of all areas strictly equals the total system power consumption, achieving a balanced and reasonable allocation. Simultaneously, the system adjusts the sensor acquisition density of each area according to the acquisition priority. Areas corresponding to risk indicator values ​​have their sampling frequency increased, transmission cycles shortened, and data fusion frequency increased. Regular areas maintain their basic frequency operation, reducing overall energy consumption while ensuring the reliability of early warnings.

[0079] Furthermore, after the resource allocation scheme is generated, the system sends the configuration instructions to the downhole sensor network through the wireless communication link to complete the unified configuration of sampling frequency, working mode, sleep cycle and transmission strategy. This ultimately forms an optimized multi-source information acquisition mode that enhances monitoring in key areas and operates with low power consumption in regular areas, thereby achieving a dual improvement in the reliability of safety early warning and the energy economy of the system.

[0080] For example, based on the Level 2 early warning results, the risk weight of area 4 downhole is set to 0.85, and the risk weight of area 5 (conventional) is set to 0.15, with a total coefficient of 0.85 + 0.15 = 1.00. The total system power budget is 1000mW, or 1W. Based on the measured total power consumption of the sensor network during normal operation, according to the allocation rules, the power quota for area 4 is 1000 × 0.85 = 850mW, and the power quota for area 5 is 1000 × 0.15 = 150mW. Simultaneously, the data acquisition density for area 4 is increased to 5 times per minute, while area 5 remains at 1 time per minute. Data completeness is defined as the ratio of the number of valid data points actually successfully acquired to the theoretically required number of data points, reflecting data loss; power utilization is defined as the ratio of the actual power consumed in high-risk areas to the total power budget, reflecting the rationality of power allocation. Performance evaluation showed that the data integrity rate was 95.2% and the power utilization rate was 85.0% (850mW / 1000mW in high-risk areas) under the current mode, representing an 8.3% improvement over the previous cycle. The data fusion frequency of the gas and temperature sensors in area 4 was ultimately increased to once every 10 seconds to determine the optimized acquisition mode.

[0081] In summary, this invention discloses a method for early warning of mine production safety. Through multi-source data cleaning and normalization, it effectively improves data continuity and reliability; through multi-parameter correlation analysis, it identifies multiple risk superposition states, significantly reducing false alarms and missed alarms; through adaptive sampling, dynamic weight adjustment, and sensor calibration, it greatly improves the accuracy and real-time performance of early warnings; and through intelligent power consumption allocation, it optimizes system energy consumption while ensuring safety monitoring effectiveness. Therefore, it effectively solves the technical problems of low early warning accuracy, poor real-time performance, fixed sampling modes, and difficulty in identifying multiple risks in traditional methods, comprehensively improving the level of intelligent monitoring and early warning for mine production safety.

[0082] The second embodiment of the present invention provides a mine production safety early warning system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described in any of the above-mentioned embodiments.

[0083] It should be noted that the mine production safety early warning system provided in this embodiment of the invention is used to execute all the process steps of the mine production safety early warning method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0084] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0085] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for early warning of mine production safety, characterized in that, include: Collect raw environmental data from multiple sources, clean the raw environmental data from multiple sources, and obtain a preliminary cleaned dataset; The initial cleaned dataset is smoothed by mean to obtain a continuously smoothed dataset. Missing values ​​in the continuously smoothed dataset are then filled by interpolation to obtain a standard cleaned dataset. Data consistency verification is performed on the standard cleaned dataset to obtain logical correlation results. Based on the logical correlation results, outlier screening is performed on the standard cleaned dataset to obtain multiple risk superposition identifiers. Based on the multiple risk overlay identifiers, a risk identifier value is calculated according to a preset initial weight. If the risk identifier value exceeds a preset risk threshold, the sampling frequency of the corresponding region of the risk identifier value is increased to obtain refined risk data. The volatility characteristics of the detailed risk data are analyzed to obtain volatility detection results. The preset initial weights are updated based on the volatility detection results to obtain a comprehensive risk assessment score. If the comprehensive risk assessment score is higher than the preset warning threshold, a regional warning signal is generated, and the sensor acquisition deviation is calibrated based on the regional warning signal to generate a real-time warning result. Based on the real-time early warning results, the acquisition density and power consumption allocation are adjusted to determine the optimized multi-source information acquisition mode.

2. The mine production safety early warning method according to claim 1, characterized in that, The process involves collecting raw environmental data from multiple sources, cleaning the raw environmental data to obtain a preliminary cleaned dataset, including: Raw environmental data from multiple sources is collected through a sensor network deployed underground. This raw environmental data includes gas concentration, dust density, and temperature. The original multi-source environmental data is filtered for noise using a median filter with a fixed window size to obtain a denoised dataset. For the denoised dataset, abnormal fluctuation points are removed by calculating the data distribution interval using the 3σ statistical criterion, resulting in a preliminary cleaned dataset.

3. The mine production safety early warning method according to claim 1, characterized in that, The process of smoothing the initial cleaned dataset by mean to obtain a continuously smoothed dataset, and then imputing missing values ​​in the continuously smoothed dataset to obtain a standard cleaned dataset includes: The initial cleaned dataset is averaged point by point according to a preset time window to complete the data smoothing process and obtain a continuous smoothed dataset. Traverse the continuous smooth dataset to identify and locate missing data points, and obtain the missing point information; The missing point information is processed by linear interpolation to obtain a standard cleaned dataset.

4. The mine production safety early warning method according to claim 2, characterized in that, The process involves performing data consistency verification on the standard cleaned dataset to obtain logical correlation results, and then filtering out anomalies based on these results to obtain multiple risk overlay identifiers, including: For the standard cleaning dataset, the correlation between the gas concentration and the dust density was analyzed using the Pearson correlation coefficient to obtain logistic correlation results; If the logical correlation result meets the preset correlation threshold, the feature points with synchronous abnormal increase are selected from the standard cleaned dataset according to the preset fluctuation threshold to obtain the abnormal synchronous feature points; For the abnormal synchronization feature points, risk levels are marked by combining preset business rules to obtain multiple risk superposition identifiers.

5. The mine production safety early warning method according to claim 2, characterized in that, The process of calculating risk identifier values ​​based on the multiple risk overlay identifiers according to preset initial weights, and increasing the sampling frequency of the corresponding region of the risk identifier value to obtain refined risk data if the risk identifier value exceeds a preset risk threshold, includes: Based on the aforementioned multiple risk superposition indicators, the risk indicator value is obtained by weighting the gas concentration, the dust density, and the temperature according to a preset initial weight. The risk identification value is compared with a preset risk threshold. If the risk identification value exceeds the preset risk threshold, the deviation ratio between the risk identification value and the average of the pre-stored historical data is calculated. Based on the deviation ratio, the sensor sampling frequency in the area corresponding to the risk identification value is dynamically increased to generate a high-frequency acquisition strategy; Data is collected according to the high-frequency acquisition strategy to obtain detailed risk data with high temporal resolution.

6. The mine production safety early warning method according to claim 5, characterized in that, The process involves performing volatility characteristic analysis on the refined risk data to obtain volatility detection results, updating the preset initial weights based on the volatility detection results, and obtaining a comprehensive risk assessment score, including: For the aforementioned detailed risk data, time series analysis is used to calculate short-term volatility characteristics and obtain volatility characteristic parameters; The volatility characteristic parameters are compared with a preset volatility threshold to obtain the volatility detection result; The preset initial weights are dynamically adjusted based on the volatility detection results to obtain optimized weights; Unify the sampling frequency, physical unit, and timestamp format of the detailed risk data to achieve format standardization and obtain standard format risk data; Based on the standard format risk data, the gas concentration, dust density, and temperature are weighted according to the optimized weights to obtain a comprehensive risk assessment score.

7. The mine production safety early warning method according to claim 1, characterized in that, If the comprehensive risk assessment score is higher than a preset warning threshold, a regional early warning signal is generated. Based on the regional early warning signal, the sensor acquisition deviation is calibrated, and a real-time early warning result is generated, including: If the comprehensive risk assessment score is higher than the preset warning threshold, a high-priority early warning signal for the corresponding downhole area is generated to obtain the regional early warning signal. Based on the aforementioned regional early warning signal, the sensor's acquisition deviation is corrected through adaptive sensitivity calibration and environmental compensation to obtain calibrated measured data. The calibration data is compared and analyzed with pre-stored historical anomaly events to obtain real-time early warning results.

8. The mine production safety early warning method according to claim 1, characterized in that, The step of adjusting the acquisition density and power consumption allocation based on the real-time early warning results to determine the optimized multi-source information acquisition mode includes: Based on the risk level of the real-time early warning results, the data collection priorities for different areas are divided to obtain a collection priority sequence; The sensor acquisition density and energy consumption quota are allocated according to the acquisition priority sequence to obtain a resource allocation scheme; Configure the sensor network according to the resource allocation scheme and determine the optimized multi-source information acquisition mode.

9. A mine production safety early warning system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the method described in any one of claims 1 to 8.

Images