A method for predicting a preventive risk index of a downhole air door

Abstract

This invention relates to the field of downhole ventilation door risk prediction technology, and discloses a method for predicting the preventive risk index of downhole ventilation doors. The method collects operational status data and environmental condition data of downhole ventilation doors, and obtains the historical risk index for each door. For each door, a reference door set is determined based on the historical risk index, and the risk sensitivity of each door at preset monitoring points is calculated. For each preset monitoring point, a preliminary risk correlation is calculated, and optimized using the door's unique identifier and risk sensitivity to obtain an optimized risk correlation. By analyzing the difference between the reference door set and the risk index distribution of all ventilation doors, the influence value of non-risk factors at the preset monitoring points is calculated, and then key monitoring points are selected based on this value and the optimized risk correlation. A risk prediction model is established based on the parameter data of the key monitoring points to predict the risk index of the target ventilation door.

Description

Technical Field

[0001] This invention relates to the field of downhole ventilation door risk prediction technology, specifically a method for predicting the preventive risk index of downhole ventilation doors. Background Technology

[0002] In mine production operations, underground air doors play a crucial role as a key component of the mine ventilation system. Just like the human respiratory system, the mine ventilation system continuously supplies fresh air to the underground working environment while expelling harmful gases and dust. Underground air doors are the core device in this system that controls airflow.

[0003] Underground ventilation doors precisely control the direction of airflow, guiding it along pre-designed routes through mine roadways. This ensures a sufficient and suitable supply of fresh air to all working faces, chambers, and other operational areas, meeting the air quality requirements for miners' normal breathing and production operations. For example, in a coal mine operation, the proper placement of underground ventilation doors allowed fresh air to reach the coal face smoothly, providing excellent ventilation conditions and ensuring efficient coal mining operations.

[0004] Underground ventilation doors can also effectively regulate airflow. Depending on the actual needs of different work areas, such as the tunneling face requiring a larger airflow to ensure ventilation due to high gas emissions and heat dissipation, while the ventilation resistance in transport roadways is lower, allowing for a smaller airflow. By adjusting the opening degree of the ventilation doors, airflow can be flexibly distributed, optimizing the operation of the ventilation system, improving ventilation efficiency, and reducing ventilation energy consumption.

[0005] Underground ventilation doors play a crucial role in ensuring a stable underground ventilation environment. They isolate airflow from different areas, prevent fresh air from mixing with exhaust air, reduce the concentration of harmful gases in the mine, and create a safe and comfortable working environment for miners. In the event of disasters such as fires or gas / coal dust explosions, underground ventilation doors can close quickly, cutting off airflow to the affected area, preventing the further spread of the disaster, buying valuable time for rescue efforts, and minimizing casualties and property damage. Therefore, the stable operation of underground ventilation doors is of irreplaceable importance for ensuring safe mine production, improving production efficiency, and promoting the sustainable development of the coal industry.

[0006] In traditional underground ventilation door management, the main method for ensuring normal operation of ventilation doors relies on regular manual inspections. Inspectors must visit each ventilation door installation location at predetermined intervals to check its appearance, sealing, and ease of operation. However, this manual inspection method has several significant drawbacks. Firstly, mines are vast, with numerous ventilation doors, making manual inspections extremely time-consuming and inefficient. For example, in some large mines, it may take several days for inspectors to complete a comprehensive inspection, during which time any ventilation door malfunctions are difficult to detect and address promptly. Secondly, manual inspections rely heavily on the inspectors' experience and visual observation, making it difficult to detect potential risks and hazards, such as minor damage to the internal structure of the ventilation door or early wear and tear of components, in a timely and accurate manner.

[0007] While some risk assessment methods for underground ventilation doors have emerged with the continuous development of technology, these methods generally have certain limitations. Some risk assessment methods consider only a single factor, such as focusing solely on the service life of the ventilation door or analyzing only the impact of airflow pressure on the door, while ignoring many other key factors that may affect the safe operation of the ventilation door, such as underground humidity, temperature, geological conditions, and equipment maintenance. This one-sided assessment approach makes the assessment results unable to comprehensively and accurately reflect the true risk status of the ventilation door, leading to difficulties in effectively predicting and preventing ventilation door failures in practical applications. For example, a coal mine used a risk assessment method based solely on the service life of the ventilation door. In a sudden ventilation door failure, this assessment method failed to issue an early warning, severely impacting mine production. This failure was caused by a sudden influx of water underground, which soaked and rusted the bottom of the ventilation door, leading to structural damage—a factor that the assessment method did not consider. Summary of the Invention

[0008] The purpose of this invention is to provide a method for predicting the preventive risk index of downhole ventilation doors, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides a method for predicting the preventive risk index of downhole ventilation doors, the method comprising:

[0010] Collect operational status data and environmental condition data of underground ventilation doors, and obtain the historical risk index of each ventilation door;

[0011] For each damper, a set of reference dampers is determined based on historical risk indices, and the risk sensitivity of each damper at preset monitoring points is calculated.

[0012] For each preset monitoring point, a preliminary risk correlation degree is calculated based on the distribution of parameter values ​​of all dampers at that monitoring point and the distribution of historical risk index.

[0013] By utilizing the unique identifier and risk sensitivity of the damper, the initial risk correlation is optimized to obtain the optimized risk correlation.

[0014] By analyzing the difference between the risk index distribution of the reference damper set and the risk index distribution of all dampers, the non-risk factor impact value of each preset monitoring point is calculated.

[0015] Based on the impact values ​​of non-risk factors and the optimized risk correlation, key monitoring points are selected from all preset monitoring points;

[0016] Based on parameter data from key monitoring points, a risk prediction model is established, and the risk index of the target damper is predicted.

[0017] Preferably, the calculation of the risk sensitivity of each damper at the preset monitoring point includes:

[0018] Calculate the coefficient of variation of the parameter values ​​of the reference damper set at the preset monitoring points, and normalize the coefficient of variation to obtain the risk sensitivity benchmark value;

[0019] Obtain the historical risk index fluctuation range of the reference damper set as a risk volatility indicator;

[0020] Arrange the risk sensitivity benchmark values ​​of all dampers in order to form a benchmark sequence, and arrange the risk fluctuation indicators of all dampers in the same order to form a fluctuation sequence.

[0021] Calculate the correlation coefficient between the benchmark sequence and the volatility sequence, and use the correlation coefficient as the risk sensitivity weight;

[0022] The risk sensitivity is obtained by multiplying the risk sensitivity weight by the risk sensitivity benchmark value.

[0023] Preferably, the calculation of the preliminary risk correlation includes:

[0024] Obtain the probability distribution of parameter values ​​for all dampers at preset monitoring points, and arrange the probability values ​​in ascending order of parameter values ​​to form a parameter probability sequence;

[0025] Obtain the probability distribution of the historical risk index of all dampers, and arrange the probability values ​​in ascending order of historical risk index to form a risk probability sequence;

[0026] The alignment degree between the parameter probability sequence and the risk probability sequence is calculated using a sequence matching algorithm, and the alignment degree is then reversed to obtain the preliminary risk correlation degree.

[0027] Preferably, obtaining the optimized risk correlation includes:

[0028] The unique identifiers of the dampers are arranged in descending order of their historical risk index, forming a risk identifier sequence;

[0029] The unique identifiers of the dampers are arranged in ascending order of the parameter values ​​at the preset monitoring points to form a parameter identifier sequence.

[0030] Identify overlapping identifiers with the same position in the risk identifier sequence and parameter identifier sequence, and calculate the ratio of the number of overlapping identifiers to the total number of identifiers as the identifier matching degree;

[0031] Remove overlapping identifiers from the risk identifier sequence to form a risk residual sequence, and replace each identifier in the risk residual sequence with the risk sensitivity of the corresponding damper to form a risk sensitivity sequence;

[0032] Remove overlapping identifiers from the parameter identifier sequence to form a parameter residual sequence, and replace each identifier in the parameter residual sequence with the risk sensitivity of the corresponding damper to form a parameter sensitivity sequence;

[0033] Calculate the similarity value between the risk sensitivity sequence and the parameter sensitivity sequence as the sensitivity matching degree;

[0034] The relevance adjustment factor is calculated by combining the identifier matching degree and the sensitivity matching degree;

[0035] The optimized risk correlation is obtained by multiplying the correlation adjustment factor by the initial risk correlation.

[0036] Preferably, the calculation of the non-risk factor impact value for each preset monitoring point includes:

[0037] For each damper, obtain the probability distribution of the historical risk index of its reference damper set, and arrange the probability values ​​in ascending order of historical risk index to form a reference risk probability sequence.

[0038] The deviation between the reference risk probability sequence and the risk probability sequence is calculated using a sequence matching algorithm, which is used as the distribution deviation of the damper.

[0039] Calculate the average distribution deviation of all dampers and perform a negative correlation mapping on the average value to obtain the impact value of non-risk factors.

[0040] Preferably, the step of selecting key monitoring points from all preset monitoring points includes:

[0041] For each preset monitoring point, calculate the ratio of the optimized risk correlation degree to the impact value of non-risk factors;

[0042] The comparison values ​​are normalized to obtain the importance score of the monitoring points;

[0043] The importance score of the monitoring point is compared with a preset threshold, and the preset monitoring points with scores higher than the threshold are selected as key monitoring points.

[0044] Preferably, the establishment of the risk prediction model includes:

[0045] Collect parameter values ​​and historical risk indices of each damper at key monitoring points to form a training dataset;

[0046] A risk prediction model is constructed by processing the training dataset using a multiple regression algorithm.

[0047] Preferably, the risk index prediction for the target damper includes:

[0048] Obtain the parameter values ​​of the target damper at key monitoring points;

[0049] Input the parameter values ​​into the risk prediction model, and output the predicted risk index of the target damper.

[0050] Preferably, determining the reference damper set includes:

[0051] Set a similarity interval for the historical risk index of the target damper. This similarity interval is centered on the historical risk index of the target damper and fluctuates by a fixed offset.

[0052] Select other dampers whose historical risk indices fall within a similar range from all dampers to form a reference damper set.

[0053] Preferably, the method further includes:

[0054] Regularly update the operating status data, environmental condition data, and historical risk index of the damper, and retrain the risk prediction model.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] This invention comprehensively assesses the operational status and environmental conditions of underground ventilation doors through multi-dimensional data collection, providing a solid information foundation for accurate risk prediction. Traditional methods often focus on only one or a few factors, making it difficult to comprehensively evaluate ventilation door risks. This invention, however, not only collects operational status data of underground ventilation doors, such as opening and closing frequency, operating time, and component wear, but also gathers environmental condition data, such as underground humidity, temperature, gas concentration, and geological conditions. This rich data provides a more realistic and comprehensive reflection of the actual conditions of the ventilation doors, offering robust data support for subsequent risk analysis.

[0057] Introducing the concepts of a reference damper set and risk sensitivity allows for a more accurate analysis of the relationship between individual dampers and the overall system, thus determining their risk level. In actual mining environments, the operational status of each damper is both interconnected and differs. This invention provides an effective reference standard for risk assessment of individual dampers by determining the reference damper set based on historical risk indices. Simultaneously, calculating the risk sensitivity of each damper at preset monitoring points makes the assessment of damper risk more detailed and precise. For example, in a certain mine, risk sensitivity analysis revealed that a particular damper has a high risk sensitivity when humidity changes. This suggests that workers should pay close attention to this damper when humidity is abnormal in the area and take preventative measures in advance.

[0058] This invention optimizes the calculation of risk correlation, making the results more closely reflect the actual risk situation. Traditional risk correlation calculations may overlook the correlation of some key factors or data, leading to deviations between the assessment results and the actual risk. This invention optimizes the preliminary risk correlation by utilizing the unique identifier of the damper and its risk sensitivity, fully considering the characteristics of different dampers and the close relationship between the parameters of each monitoring point and the risk. This results in a more accurate calculation of the risk correlation, more realistically reflecting the actual risks faced by the damper, and providing a more reliable basis for risk management. Attached Figure Description

[0059] Figure 1 This is a schematic diagram illustrating the working principle of the method for predicting the preventive risk index of downhole ventilation doors according to the present invention.

[0060] Figure 2 A flowchart for calculating the risk sensitivity of each damper at a preset monitoring point;

[0061] Figure 3 A flowchart for obtaining the optimized risk correlation;

[0062] Figure 4 A bar chart comparing the correlation between the risk of underground ventilation doors before and after optimization. Detailed Implementation

[0063] 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.

[0064] Please see Figure 1This invention provides a method for predicting the preventive risk index of underground ventilation doors. The method involves comprehensively analyzing the operating status data, environmental condition data, and historical risk indices of the underground ventilation doors to achieve accurate and automated risk prediction. The core of this method lies in using a data-driven approach to identify key monitoring points and construct a predictive model, thereby improving the reliability of risk assessment. The method includes multiple stages: data acquisition, reference set determination, risk sensitivity calculation, risk correlation analysis, non-risk factor assessment, key monitoring point selection, model building, and prediction output. The method relies on the unique identifier and historical data of the underground ventilation doors to ensure data consistency and traceability at each step. The method reduces noise interference and improves prediction accuracy through optimized processing. Implementation of this method requires the configuration of corresponding data acquisition equipment and computing units to process real-time monitoring data.

[0065] Example 1: See Figure 2 The prediction method for the preventive risk index of downhole ventilation doors involves calculating the coefficient of variation (COP) of parameter values ​​for a reference set of ventilation doors at preset monitoring points when calculating the risk sensitivity of each ventilation door at these points. The COP is the ratio of the standard deviation to the mean, used to measure the dispersion of parameter values. The COP is then normalized to a range of zero to one, yielding a baseline risk sensitivity value. The historical risk index fluctuation range of the reference set of ventilation doors is then obtained, calculated as the difference between the maximum and minimum historical risk index values. This fluctuation range serves as a risk volatility indicator. The risk sensitivity baselines for all downhole ventilation doors are then calculated. The baseline values ​​are arranged in sequence to form a baseline sequence. The sequence is based on the unique identifier or timestamp of the downhole ventilation door. The risk fluctuation indicators of all downhole ventilation doors are arranged in the same order to form a fluctuation sequence, ensuring that the sequence correspondence is consistent. The correlation coefficient between the baseline sequence and the fluctuation sequence is calculated. The correlation coefficient is the Pearson correlation coefficient, which measures the linear relationship between the two sequences. The correlation coefficient is used as the risk sensitivity weight. The risk sensitivity weight represents the correlation strength between the baseline sequence and the fluctuation sequence. The risk sensitivity weight is multiplied by the risk sensitivity baseline value to obtain the risk sensitivity. The risk sensitivity comprehensively reflects the interaction between parameter variation and risk fluctuation.

[0066] The prediction method for the preventive risk index of underground ventilation doors, when calculating the preliminary risk correlation, is based on probability distribution and sequence matching. It obtains the probability distribution of parameter values ​​for all underground ventilation doors at preset monitoring points. The probability distribution is constructed using kernel density estimation, representing the likelihood of parameter values ​​occurring. The probability values ​​are arranged in ascending order to form a parameter probability sequence, where each element represents a cumulative probability. The method also obtains the probability distribution of historical risk indices for all underground ventilation doors, derived from historical data statistics. The probability values ​​are arranged in ascending order to form a risk probability sequence, which has the same length as the parameter probability sequence. A sequence matching algorithm is used to calculate the alignment between the parameter probability sequence and the risk probability sequence. This algorithm employs dynamic time warping, and the alignment degree measures the shape similarity between the two sequences. An inverse transformation is performed on the alignment degree, achieved by taking the reciprocal, to obtain the preliminary risk correlation. A higher preliminary risk correlation value indicates a stronger correlation between the monitoring point and the risk.

[0067] Calculating the coefficient of variation (COP) of parameter values ​​for the reference damper set at preset monitoring points requires first collecting parameter values ​​for each damper in the reference damper set at designated monitoring points. These parameter values ​​are continuously collected time-series data. The arithmetic mean of these parameter values ​​is calculated, and then the deviation of each parameter value from the mean is calculated. The sum of squares of the deviations divided by the number of data points minus one yields the variance. The square root of the variance is the standard deviation. The standard deviation divided by the mean yields the COP. The COP is a dimensionless value, which can eliminate the influence of different dimensions and facilitates comparison between different monitoring points. The COP is normalized using the min-max scaling method. The minimum and maximum COP values ​​among all monitoring points are found. Each COP is subtracted from the minimum value and then divided by the difference between the maximum and minimum values, so that the normalized risk sensitivity benchmark value falls within the range of zero to one. The closer the risk sensitivity benchmark value is to one, the greater the dispersion of the parameter values ​​at the monitoring points. To obtain the fluctuation range of the historical risk index of the reference damper set, it is necessary to query the historical database and extract the historical risk index of each damper in the reference damper set within the most recent statistical period. The historical risk index is usually a standardized value. Find the maximum and minimum values ​​among these historical risk indices, and subtract the minimum value from the maximum value to obtain the fluctuation range. The fluctuation range, as an absolute value, reflects the magnitude of change in the overall risk level of the reference damper set. A larger fluctuation range means that the risk status of the dampers in the reference damper set is unstable, while a smaller fluctuation range indicates that the risk status is relatively stable. The risk fluctuation indicator directly uses this fluctuation range value without further processing, retaining its original dimensional information.

[0068] All risk sensitivity benchmark values ​​of all downhole ventilation doors are arranged sequentially to form a benchmark sequence. The order is based on the unique identifier of each downhole ventilation door, which is arranged in ascending order according to the door's installation location or numbering rules. Each downhole ventilation door corresponds to a risk sensitivity benchmark value. These risk sensitivity benchmark values ​​are placed into a list according to the order of their unique identifiers to form a benchmark sequence. The benchmark sequence is an ordered list of values, the length of which is equal to the total number of downhole ventilation doors involved in the calculation. The risk fluctuation indicators of all downhole ventilation doors are arranged in the same order to form a fluctuation sequence. The order of the fluctuation sequence must be completely consistent with the benchmark sequence. That is, the risk fluctuation indicator of the first ventilation door corresponds to the first value of the benchmark sequence, the risk fluctuation indicator of the second ventilation door corresponds to the second value of the benchmark sequence, and so on, ensuring that the correspondence between the two sequences is correct. The correlation coefficient between the benchmark series and the volatility series is calculated using the Pearson correlation coefficient formula. The Pearson correlation coefficient measures the degree of linear correlation between two variables, and its value ranges from negative one to positive one. The calculation process requires first calculating the covariance of the two series, and then calculating the standard deviation of each series. The covariance is divided by the product of the two standard deviations to obtain the Pearson correlation coefficient. A positive Pearson correlation coefficient indicates a positive correlation between the two series, a negative one indicates a negative correlation, and zero indicates no linear correlation. The calculated Pearson correlation coefficient is used as the risk sensitivity weight, which is a value between negative one and positive one. In subsequent calculations, it is necessary to ensure that the weight is non-negative; if a negative value occurs, the absolute value is taken.

[0069] The risk sensitivity is obtained by multiplying the risk sensitivity weight by the risk sensitivity benchmark value. This multiplication is performed element-wise. The risk sensitivity benchmark value for each airlock is multiplied by a uniform risk sensitivity weight to obtain the airlock's risk sensitivity at a specific monitoring point. Risk sensitivity is a comprehensive indicator that considers both the variability of the parameter value itself and the correlation strength between parameter variation and risk fluctuations. A higher risk sensitivity value indicates a more significant impact of parameter changes on the risk index at that monitoring point. The risk sensitivity value will become an important input parameter for subsequent optimization of risk correlation. The probability distribution of parameter values ​​for all downhole airlocks at preset monitoring points is obtained using kernel density estimation. Kernel density estimation is a non-parametric probability density estimation method that does not require assumptions about data distribution. The set of parameter values ​​at each monitoring point is used as input data. A suitable kernel function, such as a Gaussian kernel function, is selected, and a bandwidth parameter is set. A kernel function is applied to each data point, and all kernel functions are superimposed to obtain a smooth probability density curve. The probability density value corresponding to each parameter value is read from the probability density curve. The probability density value represents the relative likelihood of the parameter value occurring. The probability values ​​are arranged in ascending order to form a parameter probability sequence. After the parameter values ​​are sorted, the corresponding probability density values ​​are also rearranged. The elements in the parameter probability sequence are probability density values, not the original parameter values. The length of the sequence is equal to the number of parameter values. The parameter probability sequence reflects an ordered view of the distribution of parameter values, which is convenient for matching and comparison with the risk probability sequence. The probability distribution of the historical risk index of all downhole ventilation doors is also obtained using the kernel density estimation method. The processing is completely consistent with the process of obtaining the probability distribution of parameter values. The historical risk index is used as input data, and the probability density function of the risk index is obtained through kernel density estimation.

[0070] The probability values ​​are arranged in ascending order of historical risk indices to form a risk probability sequence. The generation method of the risk probability sequence is similar to that of the parametric probability sequence; both involve sorting the index values ​​and then arranging their corresponding probability density values. The length of the risk probability sequence must be consistent with that of the parametric probability sequence. If the number of risk factors differs, interpolation or truncation is required to ensure the two sequences are of equal length. The risk probability sequence characterizes the ordered nature of the risk index distribution. A sequence matching algorithm is used to calculate the alignment between the parametric and risk probability sequences. The dynamic time warping algorithm can handle the similarity matching problem between two time sequences of equal or unequal length. It uses dynamic programming to find an optimal curved path that minimizes the cumulative distance between the two sequences. Taking the parametric and risk probability sequences as input, the dynamic time warping distance between them is calculated. The smaller the dynamic time warping distance, the higher the alignment between the two sequences. The alignment is quantified as the reciprocal of the dynamic time warping distance, so that a larger alignment value indicates a more similar sequence. The alignment degree is reversed by taking the reciprocal. The dynamic time warp distance is a non-negative value. The smaller the distance, the higher the similarity. Taking the reciprocal makes the value larger, indicating a higher similarity. This reciprocal value is the preliminary risk correlation degree. The preliminary risk correlation degree is a dimensionless value. Its magnitude directly reflects the morphological similarity between the distribution of monitoring point parameters and the distribution of risk index. The closer the preliminary risk correlation degree is to one, the stronger the correlation between the monitoring point and the risk. The closer the value is to zero, the weaker the correlation. The preliminary risk correlation degree will be used as a basic indicator in the subsequent optimization process.

[0071] Example 2: See Figure 3The prediction method for the preventive risk index of underground ventilation doors utilizes the unique identifier and risk sensitivity of underground ventilation doors for optimization when obtaining the optimized risk correlation. The unique identifiers of underground ventilation doors are arranged in descending order of their historical risk index, forming a risk identifier sequence. The unique identifier is the code or ID number of the underground ventilation door. The unique identifiers of underground ventilation doors are then arranged in ascending order of their parameter values ​​at preset monitoring points, forming a parameter identifier sequence. The parameter value sorting is based on numerical size or tiered processing. Overlapping identifiers in the risk identifier sequence and parameter identifier sequence are identified. Overlapping identifiers indicate underground ventilation doors that are in the same position under both sortings. The ratio of the number of overlapping identifiers to the total number of identifiers is calculated as the identifier matching degree, reflecting the degree of sorting consistency. Overlapping identifiers are removed from the risk identifier sequence, forming a residual risk sequence. The residual risk sequence contains underground ventilation door identifiers with inconsistent sorting. Each identifier in the residual risk sequence... The identifiers are replaced with the risk sensitivities of the corresponding downhole ventilation doors, forming a risk sensitivities sequence. The elements of the risk sensitivities sequence are numerical data. Overlapping identifiers are removed from the parameter identifier sequence to form a parameter residual sequence. The parameter residual sequence is processed similarly to the risk residual sequence, with each identifier in the parameter residual sequence replaced with the risk sensitivities of the corresponding downhole ventilation door, forming a parameter sensitivities sequence. The parameter sensitivities sequence and the risk sensitivities sequence have the same length. The similarity value between the risk sensitivities sequence and the parameter sensitivities sequence is calculated using cosine similarity as the sensitivity matching degree. The sensitivity matching degree measures the similarity of the sensitivity distribution of the remaining identifiers. Combining the identifier matching degree and the sensitivity matching degree, a correlation adjustment factor is calculated. The correlation adjustment factor is obtained through weighted average calculation. The correlation adjustment factor is multiplied by the preliminary risk correlation degree to obtain the optimized risk correlation degree. The optimized risk correlation degree integrates identifier consistency and sensitivity information to reduce the impact of individual outliers.

[0072] To form a risk identification sequence, it is necessary to first obtain a list of unique identifiers for all underground ventilation doors involved in the calculation. A unique identifier is a string or numeric code with unique characteristics. These identifiers are sorted according to the historical risk index value of each underground ventilation door in the most recent assessment period, arranged in descending order from highest to lowest. Correspondingly, the unique identifiers of these underground ventilation doors are arranged in the same order to generate an ordered list of identifiers, i.e., the risk identification sequence. The first element in the risk identification sequence is the unique identifier of the underground ventilation door with the highest historical risk index, and the last element is the unique identifier of the underground ventilation door with the lowest historical risk index. The sequence length is equal to the total number of underground ventilation doors. The order of the risk identification sequence reflects the relative historical risk level of the underground ventilation doors. The process of forming the parameter identifier sequence is similar to that of the risk identifier sequence, but the sorting is based on the parameter values ​​of the downhole air doors at specific preset monitoring points. The parameter values ​​are continuous numerical data, and they are arranged in ascending order from smallest to largest. Correspondingly, the unique identifiers of the downhole air doors are arranged in order of parameter values ​​to generate the parameter identifier sequence. The first element in the parameter identifier sequence is the unique identifier of the downhole air door with the smallest parameter value, and the last element is the unique identifier of the downhole air door with the largest parameter value. The sorting of the parameter identifier sequence reflects the relative magnitude of the parameter values ​​of the downhole air doors at the monitoring points.

[0073] Identifying overlapping identifiers in the risk identifier sequence and parameter identifier sequence requires comparing the two sequences position by position. The identifier at the first position of the risk identifier sequence is compared with the identifier at the first position of the parameter identifier sequence. If the two identifiers are the same, they are recorded as an overlapping identifier. The identifier at the second position of the risk identifier sequence is compared with the identifier at the second position of the parameter identifier sequence, and so on until the end of the sequence. An overlapping identifier is a unique identifier of the same downhole ventilation door that is in the same order in the two sequences. The number of overlapping identifiers is obtained by counting. The total number of identifiers is the length of the sequence. The number of overlapping identifiers divided by the total number of identifiers gives the identifier matching degree. The identifier matching degree is a value between zero and one. The higher the identifier matching degree, the better the consistency between the results of sorting by risk and sorting by parameter value. The risk residual sequence is formed by removing overlapping identifiers from the risk identifier sequence. Each identifier in the risk identifier sequence is traversed and checked to see if it belongs to the overlapping identifier set. If it does not belong to the overlapping identifier set, the identifier is retained and placed into a new sequence in the original order. This new sequence is the risk residual sequence. The risk residual sequence contains downhole ventilation door identifiers that are in different positions according to risk sorting and parameter value sorting. The risk residual sequence maintains the relative order of non-overlapping identifiers in the original risk identifier sequence.

[0074] Each identifier in the remaining risk sequence is replaced with the risk sensitivity value of the corresponding downhole ventilation door to form a risk sensitivity sequence. The risk sensitivity value is a pre-calculated attribute value for each downhole ventilation door. For each identifier in the remaining risk sequence, the risk sensitivity value of the corresponding downhole ventilation door is found, and this value replaces the original identifier, generating a pure numerical sequence, i.e., the risk sensitivity sequence. The order of the values ​​in the risk sensitivity sequence is exactly the same as the order of the identifiers in the remaining risk sequence. Overlapping identifiers are removed from the parameter identifier sequence to form the parameter residual sequence. The processing method is completely symmetrical to the formation of the remaining risk sequence. Each identifier in the parameter identifier sequence is traversed, overlapping identifiers are removed, and non-overlapping identifiers are retained while maintaining their original order, generating the parameter residual sequence. The parameter residual sequence contains those identifiers in the parameter identifier sequence that do not appear in the same position. Each identifier in the parameter residual sequence is replaced with the risk sensitivity of the corresponding downhole ventilation door to form a parameter sensitivity sequence. For each identifier in the parameter residual sequence, the risk sensitivity value of the corresponding downhole ventilation door is found and replaced to generate a parameter sensitivity sequence. The parameter sensitivity sequence and the risk sensitivity sequence have the same length because both residual sequences are generated from non-overlapping identifiers and have the same number.

[0075] The similarity between the risk sensitivity sequence and the parameter sensitivity sequence is calculated using the cosine similarity algorithm. Cosine similarity measures the similarity by calculating the cosine of the angle between two vectors. The risk sensitivity sequence is considered as vector A, and the parameter sensitivity sequence as vector B. The dot product of vectors A and B is calculated, and then the magnitudes of vectors A and B are calculated separately. The dot product divided by the product of the two magnitudes yields the cosine similarity value. The cosine similarity value ranges from -1 to +1; the closer the value is to one, the more consistent the directions of the two vectors. The sensitivity matching score directly uses this cosine similarity value. A correlation adjustment factor is calculated by combining the identifier matching score and the sensitivity matching score. The correlation adjustment factor is a weighted average of the identifier matching score and the sensitivity matching score. One weight coefficient is assigned to the identifier matching score, and another weight coefficient is assigned to the sensitivity matching score. The sum of the two weight coefficients is one. The correlation adjustment factor is obtained by multiplying the identifier matching score by its weight and adding the sensitivity matching score multiplied by its weight. The correlation adjustment factor is a value between zero and one. The optimized risk correlation is obtained by multiplying the correlation adjustment factor by the preliminary risk correlation. The preliminary risk correlation is a value calculated in the previous steps that reflects the degree of correlation between the monitoring point and the risk basis. The multiplication operation is a scalar multiplication. The correlation adjustment factor is used as an adjustment coefficient to scale the preliminary risk correlation. The optimized risk correlation retains the basic information of the preliminary risk correlation and incorporates the characteristics of the consistency of the identifier ranking and the distribution of risk sensitivity. The optimized risk correlation has better robustness and representativeness.

[0076] See Figure 4This chart visually presents the comparison between the initial and optimized risk correlation of downhole ventilation doors using a bar chart format. The horizontal axis represents the ventilation door serial number, and the vertical axis represents the risk correlation. Orange bars represent the initial risk correlation, while blue bars represent the correlation optimized by unique ventilation door identifiers and risk sensitivity. By arranging the identifier sequence of historical risk indices and monitoring point parameter values ​​for ventilation doors, the identifier matching degree and sensitivity matching degree are calculated, thereby obtaining the correlation adjustment factor. This factor is then multiplied by the initial risk correlation to obtain the optimized risk correlation. The chart clearly shows the difference in correlation before and after optimization, demonstrating the corrective effect of this invention on the initial risk correlation. This provides a more accurate basis for subsequent key monitoring point selection and risk prediction model establishment, solving the evaluation bias problem caused by neglecting the individual characteristics of ventilation doors and data correlation in traditional risk correlation calculations, ultimately improving the accuracy of downhole ventilation door risk prediction.

[0077] Example 3: Prediction Method for Preventive Risk Index of Downhole Air Doors. When calculating the impact value of non-risk factors at each preset monitoring point, based on the distribution difference between the reference air door set and all air doors, for each downhole air door, the probability distribution of its historical risk index from the reference air door set is obtained. This probability distribution is constructed using an empirical distribution function, and the probability values ​​are arranged in ascending order of historical risk index to form a reference risk probability sequence. This reference risk probability sequence represents the risk distribution of similar air doors. The probability distribution of the historical risk index of all downhole air doors is then obtained to form a risk probability sequence, which represents the risk distribution of the entire air door group. The deviation between the reference risk probability sequence and the risk probability sequence is calculated using a sequence matching algorithm. The sequence matching algorithm employs the Friesian distance algorithm. The deviation measures the shape difference between the two sequences and serves as the distribution deviation of the downhole ventilation doors. The distribution deviation indicates the degree of inconsistency between the risk distribution of the reference set and that of all ventilation doors. The average distribution deviation of all downhole ventilation doors is calculated using an arithmetic mean, reflecting the overall deviation level at the monitoring points. A negative correlation mapping is performed on the average, achieved through a linear transformation, to obtain the influence value of non-risk factors. A higher non-risk factor influence value indicates that the monitoring point is less affected by non-risk factors.

[0078] To obtain the probability distribution of the historical risk index of each underground ventilation door's reference set, it's necessary to first determine the members of the reference set. This set consists of underground ventilation doors with similar historical risk indices. Historical risk index data for these doors over a specific time period is collected, and a probability distribution is constructed using an empirical distribution function (EPF). The EPF is a step function that jumps at each data point, with the function value equal to or less than the proportion of data points. A series of discrete probability values ​​are read from the EPF, and these values ​​are arranged in ascending order of historical risk index values ​​to form a reference risk probability sequence. The length of the reference risk probability sequence is equal to the number of selected discrete points. The same method is used to obtain the probability distribution of the historical risk index for all underground ventilation doors. Historical risk index data for all underground ventilation doors is collected, and the EPF is applied to obtain the probability distribution. The probability values ​​are then arranged in ascending order of historical risk index values ​​to form a risk probability sequence. The length of the risk probability sequence is consistent with the length of the reference risk probability sequence, which can be achieved by selecting the same number of discrete points. The risk probability sequence provides a complete picture of the risk distribution of the entire ventilation door group.

[0079] The Fraser distance algorithm is used to calculate the deviation between the reference risk probability sequence and the risk probability sequence. The Fraser distance measures the similarity between two curves, taking into account the correspondence between points and the continuous changes along the path. The reference risk probability sequence is regarded as curve A, and the risk probability sequence is regarded as curve B. The calculation of the Fraser distance requires finding a way to reparameterize curves A and B so that the correspondence between points on the two curves minimizes the maximum distance. This minimized maximum distance is the Fraser distance value, which is used as the deviation between the two sequences.

[0080] The formula for calculating the Frescher distance is:

[0081]

[0082] in: This represents the Frescher distance between the reference risk probability sequence and the risk probability sequence. Let B represent the parameterized curve of the reference risk probability sequence. and It is a reparameterization function used to align points on two curves. It is the Euclidean distance function between two points. The curve parameters take values ​​between zero and one. This indicates taking the infimum of all reparameterized combinations. Indicates in the parameter The maximum distance value is taken within the range of values.

[0083] The deviation value is directly calculated using the Fraser distance. A larger deviation value indicates a greater difference between the risk distribution of the reference set of airlocks and the overall risk distribution of all airlocks. This deviation value serves as the distribution deviation of the downhole airlocks, quantifying the degree of dissimilarity between the reference set and the overall group in terms of risk distribution patterns. Calculating the average distribution deviation of all downhole airlocks requires collecting the distribution deviation values ​​for each downhole airlock. These distribution deviation values ​​form a dataset. The arithmetic mean of this dataset is calculated by summing all distribution deviations and dividing by the total number of downhole airlocks. This average value reflects the central tendency of the distribution deviation between the reference set and the overall group of airlocks at a specific monitoring point.

[0084] Negative correlation mapping of the average value is achieved through a linear transformation. This negative correlation mapping converts the average distribution deviation into a value proportional to the influence of non-risk factors. The mapping formula is as follows: When the average distribution deviation is large, the impact value of non-risk factors is small, indicating that the monitoring point is more affected by non-risk factors. Conversely, when the average distribution deviation is small, the impact value of non-risk factors is large, indicating that the monitoring point is less affected by non-risk factors. The impact value of non-risk factors is a value between zero and one. The closer the value is to one, the less affected the monitoring point is by non-risk factors; the closer the value is to zero, the more affected the monitoring point is by non-risk factors. The impact value of non-risk factors is used in subsequent critical monitoring point screening steps to help identify those monitoring points that are truly related to risk rather than being affected by external factors.

[0085] The prediction method for the underground ventilation door preventive risk index isolates non-risk factors, such as the impact of equipment aging or sudden environmental changes, through distribution deviation analysis. The method considers the overall shape of the probability sequence in deviation calculation, avoiding misjudgments caused by local fluctuations. The impact value of non-risk factors, as a quantitative indicator, enhances the robustness and accuracy of the risk prediction model. The entire calculation process is based on rigorous mathematics, with clear data processing and transformation logic at each step. The impact value of non-risk factors is ultimately combined with other indicators to select the key monitoring points with the strongest risk prediction capabilities. The application of the Fraser distance algorithm makes the measurement of distribution deviation more accurate. The Fraser distance considers the overall shape of the sequence and the correspondence between points, capturing the essential differences in probability distribution curves better than the simple Euclidean distance. The use of the empirical distribution function avoids prior assumptions about the data distribution form, enhancing the method's adaptability to various risk data distributions. The linear transformation form of the negative correlation mapping is simple and effective, ensuring that the impact value of non-risk factors has clear mathematical meaning and interpretability. The prediction method for the preventive risk index of underground ventilation doors reflects a deep understanding of the essence of data in the calculation of the impact value of non-risk factors. By comparing the distribution of the reference ventilation door set with that of all ventilation doors, it identifies external factors that may interfere with risk assessment. The impact value of non-risk factors is combined with traditional risk correlation indicators to form a more comprehensive monitoring point assessment system. This comprehensive assessment method improves the accuracy and reliability of risk prediction and provides a scientific basis for the safety management of underground ventilation doors.

[0086] The calculation process for the impact value of non-risk factors is highly repeatable and verifiable, with clear data inputs and outputs for each step, facilitating implementation and debugging in practical applications. The Fréchet distance can be calculated using standard algorithms, and the empirical distribution function is constructed based on historical data. The entire process requires no human intervention or subjective judgment, ensuring consistency and objectivity of the results. The impact value of non-risk factors, as a crucial input parameter for the risk prediction model, directly influences the selection and quality of key monitoring points. The prediction method for the preventive risk index of downhole ventilation doors enhances the model's adaptability to complex downhole environments by introducing the impact value of non-risk factors. Various external factors in the downhole environment may interfere with risk assessment; the impact value of non-risk factors effectively isolates these interferences, highlighting the true risk signals. This design makes the risk prediction model more stable and reliable in practical applications, providing accurate guidance for the preventive maintenance of downhole ventilation doors.

[0087] Example 4: Prediction Method for the Preventive Risk Index of Downhole Air Doors. When selecting key monitoring points from all preset monitoring points, the method combines the optimized risk correlation degree with the influence value of non-risk factors. For each preset monitoring point, the ratio of the optimized risk correlation degree to the influence value of non-risk factors is calculated. The ratio is obtained through division and represents the net risk correlation strength of the monitoring point. The ratio is normalized using the minimum-maximum scaling method to obtain the importance score of the monitoring point. The importance score of the monitoring point ranges from zero to one. The importance score of the monitoring point is compared with a preset threshold, which is set based on historical experience or statistical quantiles. Preset monitoring points with scores higher than the threshold are selected as key monitoring points. The key monitoring points constitute the core input set for risk prediction. The method for predicting the preventive risk index of underground ventilation doors uses data from key monitoring points when establishing a risk prediction model. This involves collecting parameter values ​​and historical risk indices for each underground ventilation door at these key monitoring points. Parameter values ​​include operating status and environmental condition data, while historical risk indices serve as label values. These form a training dataset, with parameter values ​​as feature vectors and historical risk indices as target variables. A multiple regression algorithm is used to process the training dataset, employing linear regression to construct the risk prediction model. The risk prediction model outputs an estimated risk index value.

[0088] Calculating the ratio of optimized risk correlation to the impact value of non-risk factors requires obtaining the optimized risk correlation value and the impact value of non-risk factors for each preset monitoring point. The optimized risk correlation is a correlation index after label matching and sensitivity adjustment. The impact value of non-risk factors reflects the degree of interference of the monitoring point with external factors. The ratio obtained by dividing the optimized risk correlation by the impact value of non-risk factors is called the monitoring point correlation strength ratio. The larger the monitoring point correlation strength ratio, the stronger the real correlation between the monitoring point and the risk, and the less interference from non-risk factors. The comparison value is normalized using the minimum-maximum scaling method. First, the monitoring point correlation strength ratio values ​​of all preset monitoring points are collected, and the minimum value MIN and maximum value MAX are found among these values. For the monitoring point correlation strength ratio R of each monitoring point, the normalized value is calculated using the formula. The normalized value is equal to the monitoring point correlation strength ratio R minus MIN divided by MAX minus MIN. The value obtained after normalization is called the monitoring point importance score. The monitoring point importance score maps the original ratio to the range of zero to one, which facilitates comparison between different monitoring points and threshold screening.

[0089] The preset threshold is set based on historical data analysis. Importance scores of all monitoring points in past cycles are collected, and the statistical quantiles of these scores are calculated. The 75th quantile is selected as the preset threshold, a fixed value between zero and one, used to determine whether a monitoring point is sufficiently important. The process of selecting key monitoring points involves comparing each monitoring point individually. The importance score of each monitoring point is compared to the preset threshold. Monitoring points with an importance score greater than the preset threshold are selected as key monitoring points. The set of key monitoring points represents the sensor locations or data acquisition points with the greatest risk prediction capability. When collecting parameter values ​​for each downhole ventilation door at key monitoring points, the specific location and measurement type of the key monitoring points need to be determined. Key monitoring points include ventilation door bearing temperature monitoring points, door vibration monitoring points, and airflow pressure monitoring points. The latest parameter readings for each downhole ventilation door at these key monitoring points are extracted from the data acquisition system. Parameter values ​​are usually numerical data and may require dimension unification and missing value handling. The historical risk index is obtained from the ventilation door maintenance database. The historical risk index is a risk score that has been evaluated by experts or calculated automatically. The value of the historical risk index ranges from zero to one hundred. The higher the value, the more serious the risk. Each downhole ventilation door corresponds to a historical risk index value.

[0090] The training dataset is composed of a feature matrix and label vectors. Rows in the feature matrix correspond to different underground ventilation doors, and columns correspond to different key monitoring points. Matrix elements are the parameter values ​​of the underground ventilation doors at the key monitoring points. The label vectors are the historical risk index values ​​for each underground ventilation door. The size of the training dataset depends on the number of available underground ventilation doors and data completeness. The multiple regression algorithm uses ordinary least squares linear regression. The linear regression model is defined as the risk index equal to the parameter value multiplied by the coefficient plus the intercept term. The coefficients and intercept are obtained by minimizing the sum of squared prediction errors. The mathematical expression of the linear regression model represents the linear relationship between the risk index and the parameter values ​​at multiple monitoring points. The training process of the risk prediction model uses a numerical optimization algorithm to solve for the optimal estimates of the linear regression coefficients. The trained model can output predicted risk index values ​​based on the input key monitoring point parameter values. Refer to Table 1, which shows the structure of a training dataset containing the parameter values ​​of three underground ventilation doors at three key monitoring points and their corresponding historical risk indices.

[0091] Table 1: Training Data Set Table

[0092] Unique identifier for underground ventilation doors Key monitoring point A parameter value Key monitoring point B parameter value Key monitoring point C parameter value Historical Risk Index FAN_DOOR_001 45.6 12.8 0.56 75 FAN_DOOR_002 38.9 15.3 0.62 68 FAN_DOOR_003 51.2 11.5 0.53 82

[0093] Each row in Table 1 represents a training sample, and each training sample contains complete data for one underground ventilation door. Multiple training samples form a training dataset used to train the risk prediction model. The quality of the selection of key monitoring points directly affects the performance of the risk prediction model. The calculation of the importance score of monitoring points ensures that only those monitoring points that are highly correlated with the risk and are less affected by interference are selected. Normalization processing makes the correlation strength ratio of monitoring points of different dimensions comparable. The setting of the preset threshold needs to balance the complexity of the model and the prediction accuracy. Too low a threshold will introduce too many monitoring points and increase the complexity of the model, while too high a threshold may miss important monitoring points and reduce the prediction ability.

[0094] Quality control of the training dataset includes data cleaning and outlier handling. Obvious outliers in the parameter values ​​need to be identified and corrected. The accuracy and consistency of historical risk indices need to be guaranteed through data validation. The correspondence between the feature matrix and the label vector must be accurate. The coefficients of the linear regression model have clear physical meanings: positive coefficients indicate a positive correlation between the parameter value and the risk index, while negative coefficients indicate a negative correlation. The absolute value of the coefficient reflects the degree of influence of the parameter value on the risk index. After model training, performance evaluation is required. Commonly used evaluation metrics include mean squared error and coefficient of determination. The practical application of the risk prediction model requires regular updates. When a certain amount of new ventilation door data accumulates, the model is retrained to maintain prediction accuracy. The model update cycle is determined based on the rate of data change and application requirements, typically monthly or quarterly. The prediction method for the underground ventilation door preventative risk index achieves quantitative and automated risk prediction through systematic key monitoring point selection and model construction. The key monitoring point selection process reduces data dimensionality and highlights core risk factors. The linear regression model provides an intuitive explanation of risk relationships. The construction of the training dataset needs to consider the temporal consistency of the data. The parameter values ​​and historical risk indices of the same air door should come from the same time period. The collection frequency of parameter values ​​and the evaluation cycle of historical risk indices need to be coordinated to ensure that the data can accurately reflect the air door status. The output of the risk prediction model can be used to trigger maintenance early warnings. When the predicted risk index exceeds a certain threshold, the system automatically generates a maintenance work order to guide maintenance personnel to conduct targeted inspections. This preventive maintenance mode can reduce the air door failure rate and improve the reliability of the downhole ventilation system.

[0095] The selection of key monitoring points is a dynamic process. As the downhole environment changes and the condition of ventilation doors evolves, the set of key monitoring points needs to be reassessed periodically. Changes in the importance scores of monitoring points can reflect changes in system risk characteristics, providing data support for ventilation system optimization. The implementation of the prediction method for the downhole ventilation door preventative risk index requires supporting data acquisition and storage infrastructure. The coverage density of the sensor network and the reliability of data transmission affect the quality of parameter values, while the performance and capacity of the database system determine the integrity and accessibility of historical data. The interpretability of linear regression models helps field engineers understand the causes of risk. By analyzing model coefficients, major risk factors can be identified, allowing for targeted strengthening of monitoring and management of these factors. The changing trends of model coefficients can reflect the aging patterns of ventilation doors and the impact of environmental factors.

[0096] Example 5: Prediction Method for Preventive Risk Index of Downhole Air Doors When predicting the risk index of a target air door, a trained risk prediction model is used. The target air door, numbered FAN_DOOR_20240615, is located in the south wing transport roadway at a depth of -350 meters. Its risk index for the next thirty days needs to be predicted. Parameter values ​​at key monitoring points of the target air door are obtained through a sensor network deployed on the air door itself. These key monitoring points include three locations: door shaft vibration acceleration monitoring point, sealing pressure monitoring point, and drive motor current monitoring point. The parameter value for the door shaft vibration acceleration monitoring point is the effective vibration value collected per second; the parameter value for the sealing pressure monitoring point is the average pressure recorded per minute; and the parameter value for the drive motor current monitoring point is the root mean square value of the current sampled every ten seconds. These parameter values ​​are transmitted to a ground data center via industrial Ethernet. The data format is floating-point numbers with timestamps and device identifiers. The risk prediction model is a mathematical model trained using a linear regression algorithm. The model input dimension is the parameter values ​​of the three key monitoring points, and the output is a risk index value between zero and one hundred. The model coefficients are embedded in the prediction system, with a coefficient matrix of [0.35, 1.2, -0.8] and an intercept term of 12.5. When inputting the parameter values ​​of the target damper into the risk prediction model, the data vector needs to be organized in a fixed order: the first dimension is the damper shaft vibration acceleration value, the second dimension is the sealing pressure value, and the third dimension is the drive motor current value. The data vector is standardized by subtracting the mean of the training set and dividing by the standard deviation of the training set to ensure that the input data maintains the same distribution characteristics as the training data.

[0097] Setting a similarity interval for the historical risk index of the target ventilation door requires referencing its risk records over the past twelve months. The average historical risk index of the target ventilation door FAN_DOOR_20240615 is 65 points, and the standard deviation of the twelve-month risk index is 8 points. The similarity interval is set with 65 points as the center, fluctuating by two standard deviations (16 points) above and below, forming a risk index range of 49 to 81 points. Other ventilation doors whose historical risk indices fall within the 49 to 81 point range are screened from the database of all 120 ventilation doors in the mine ventilation system. The screening results yield 15 qualified ventilation doors. These 15 ventilation doors form a reference ventilation door set, including those numbered FAN_DOOR_20240013 and FAN_DOOR_20240027. These ventilation doors have similar risk characteristics to the target ventilation door. The determination process for the reference ventilation door set considers the installation location and operating environment of the ventilation doors. All ventilation doors are located in transport roadways at the -350m to -400m level, experiencing similar geological stress conditions and ventilation loads. The reference damper set consists entirely of double-leaf, opposing steel dampers, driven by an electro-hydraulic actuator, and have been in service for three to five years. These common characteristics ensure the comparability between the reference damper set and the target damper set, avoiding reference deviations caused by structural differences or environmental variations.

[0098] Obtaining parameter values ​​for the target damper at key monitoring points requires processing real-time data streams. The sensor for the damper shaft vibration acceleration monitoring point is model XYZ-3A, with a range of 0 to 10 grams and an accuracy of 0.5%. The sealing pressure monitoring point uses a PTC-200 pressure transmitter, with a measurement range of 0 to 5 MPa. The drive motor current monitoring point acquires signals through a Hall effect sensor, outputting a 4 to 20 mA analog signal. The data acquisition system transmits monitoring data in packets every minute, with each packet containing device identification, parameter values, timestamps, and quality inspection marks. Invalid data is automatically filtered out; for example, records with vibration acceleration values ​​exceeding 10 grams or current values ​​less than zero are considered outliers and removed. The risk prediction model calculation process is automatically executed on the server side. After standardization, the input vector is multiplied by the coefficient matrix, and an intercept term is added to obtain the original output. The original output is mapped to the 0 to 100 range using the Sigmoid function, ultimately generating the predicted risk index. The entire calculation process takes less than 100 milliseconds and supports concurrent processing of prediction requests from multiple dampers. The prediction results are stored in the risk warning database and simultaneously pushed to the ventilation safety management platform. The dynamic update mechanism for the reference damper set is executed monthly. At the beginning of each month, the system automatically recalculates the historical risk index average for each damper and adjusts the similarity range based on the latest data. Newly commissioned dampers are included in the screening scope after a three-month observation period, while decommissioned or scrapped dampers are promptly removed from the database. This dynamic adjustment mechanism ensures that the reference damper set always reflects the actual operating status of the current system.

[0099] The prediction example for the target damper FAN_DOOR_20240615 demonstrates the complete processing flow. The parameter values ​​collected on a certain day were: door shaft vibration acceleration monitoring point 2.3 g, sealing pressure monitoring point 1.5 MPa, and drive motor current monitoring point 8.6 amperes. After data vector standardization, the data was input into the risk prediction model, calculating a predicted risk index of 73. This value exceeded the preset warning threshold of 60, and the system automatically generated a level-two warning notification, prompting maintenance personnel to increase the frequency of damper inspections. Verification of the prediction results relied on subsequent actual operational data. Within thirty days of the prediction being generated, the target damper did indeed exhibit accelerated wear of the sealing strip, with an actual risk index assessment of 76. The error between the predicted and actual values ​​was within the allowable range, proving the effectiveness of the prediction method. This data-driven prediction method provides a scientific basis for preventative maintenance of dampers. The selection algorithm for the reference damper set considered spatiotemporal correlation. In addition to risk index similarity, the principle of geographical proximity was also introduced, prioritizing dampers from the same mining area or adjacent roadways. In terms of time dimension, the reference and target air doors must be within the same service life stage to avoid comparing old and new equipment together. These additional conditions further enhance the representativeness of the reference air door set. The prediction method for the preventive risk index of underground air doors is demonstrated through an example of a target air door, showcasing the complete technical chain from data acquisition to result output. Real-time acquisition of parameter values, efficient calculation of the risk prediction model, and intelligent screening of the reference air door set are all closely linked to form an organic whole. The prediction results directly serve ventilation safety management, realizing a shift from passive maintenance to proactive prevention.

[0100] The risk prediction model requires continuous updates and maintenance. When the ventilation door equipment within the system undergoes technical upgrades or model changes, training data needs to be collected again and model parameters updated. The model performance monitoring module periodically evaluates the prediction accuracy, triggering a model retraining process when the error exceeds 15%. This self-improving mechanism ensures that the prediction system can adapt to changes in mine conditions. The method for determining the reference ventilation door set is flexible and scalable. For newly built mines or new ventilation door groups, transfer learning techniques can be used to borrow reference models from other mines, and a localized reference system can be established after sufficient data has been accumulated. This method solves the data cold start problem and accelerates system deployment. The technical details demonstrated in the target ventilation door prediction example reflect the practicality of the method. The selection specifications for parameter acquisition equipment, the standardization of data transmission protocols, and the optimized design of the calculation process all contribute to ensuring the stable operation of the prediction system.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the preventive risk index of downhole ventilation doors, characterized in that, The method includes: Collect operational status data and environmental condition data of underground ventilation doors, and obtain the historical risk index of each ventilation door; For each damper, a set of reference dampers is determined based on historical risk indices, and the risk sensitivity of each damper at preset monitoring points is calculated. For each preset monitoring point, a preliminary risk correlation degree is calculated based on the distribution of parameter values ​​of all dampers at that monitoring point and the distribution of historical risk index. By utilizing the unique identifier and risk sensitivity of the damper, the initial risk correlation is optimized to obtain the optimized risk correlation. By analyzing the difference between the risk index distribution of the reference damper set and the risk index distribution of all dampers, the non-risk factor impact value of each preset monitoring point is calculated. Based on the impact values ​​of non-risk factors and the optimized risk correlation, key monitoring points are selected from all preset monitoring points; Based on parameter data from key monitoring points, a risk prediction model is established, and the risk index of the target damper is predicted.

2. The method for predicting the preventive risk index of downhole ventilation doors according to claim 1, characterized in that, The calculation of the risk sensitivity of each damper at the preset monitoring point includes: Calculate the coefficient of variation of the parameter values ​​of the reference damper set at the preset monitoring points, and normalize the coefficient of variation to obtain the risk sensitivity benchmark value; Obtain the historical risk index fluctuation range of the reference damper set as a risk volatility indicator; Arrange the risk sensitivity benchmark values ​​of all dampers in order to form a benchmark sequence, and arrange the risk fluctuation indicators of all dampers in the same order to form a fluctuation sequence. Calculate the correlation coefficient between the benchmark sequence and the volatility sequence, and use the correlation coefficient as the risk sensitivity weight; The risk sensitivity is obtained by multiplying the risk sensitivity weight by the risk sensitivity benchmark value.

3. The method for predicting the preventive risk index of downhole ventilation doors according to claim 1, characterized in that, The calculation of the preliminary risk correlation includes: Obtain the probability distribution of parameter values ​​for all dampers at preset monitoring points, and arrange the probability values ​​in ascending order of parameter values ​​to form a parameter probability sequence; Obtain the probability distribution of the historical risk index of all dampers, and arrange the probability values ​​in ascending order of historical risk index to form a risk probability sequence; The alignment degree between the parameter probability sequence and the risk probability sequence is calculated using a sequence matching algorithm, and the alignment degree is then reversed to obtain the preliminary risk correlation degree.

4. The method for predicting the preventive risk index of downhole ventilation doors according to claim 1, characterized in that, The optimized risk correlation degree includes: The unique identifiers of the dampers are arranged in descending order of their historical risk index, forming a risk identifier sequence; The unique identifiers of the dampers are arranged in ascending order of the parameter values ​​at the preset monitoring points to form a parameter identifier sequence. Identify overlapping identifiers with the same position in the risk identifier sequence and parameter identifier sequence, and calculate the ratio of the number of overlapping identifiers to the total number of identifiers as the identifier matching degree; Remove overlapping identifiers from the risk identifier sequence to form a risk residual sequence, and replace each identifier in the risk residual sequence with the risk sensitivity of the corresponding damper to form a risk sensitivity sequence; Remove overlapping identifiers from the parameter identifier sequence to form a parameter residual sequence, and replace each identifier in the parameter residual sequence with the risk sensitivity of the corresponding damper to form a parameter sensitivity sequence; Calculate the similarity value between the risk sensitivity sequence and the parameter sensitivity sequence as the sensitivity matching degree; The relevance adjustment factor is calculated by combining the identifier matching degree and the sensitivity matching degree; The optimized risk correlation is obtained by multiplying the correlation adjustment factor by the initial risk correlation.

5. The method for predicting the preventive risk index of downhole ventilation doors according to claim 3, characterized in that, The calculation of the non-risk factor impact value for each preset monitoring point includes: For each damper, obtain the probability distribution of the historical risk index of its reference damper set, and arrange the probability values ​​in ascending order of historical risk index to form a reference risk probability sequence. The deviation between the reference risk probability sequence and the risk probability sequence is calculated using a sequence matching algorithm, which is used as the distribution deviation of the damper. Calculate the average distribution deviation of all dampers and perform a negative correlation mapping on the average value to obtain the impact value of non-risk factors.

6. The method for predicting the preventive risk index of downhole ventilation doors according to claim 1, characterized in that, The process of selecting key monitoring points from all preset monitoring points includes: For each preset monitoring point, calculate the ratio of the optimized risk correlation degree to the impact value of non-risk factors; The comparison values ​​are normalized to obtain the importance score of the monitoring points; The importance score of the monitoring point is compared with a preset threshold, and the preset monitoring points with scores higher than the threshold are selected as key monitoring points.

7. The method for predicting the preventive risk index of downhole ventilation doors according to claim 1, characterized in that, The establishment of the risk prediction model includes: Collect parameter values ​​and historical risk indices of each damper at key monitoring points to form a training dataset; A risk prediction model is constructed by processing the training dataset using a multiple regression algorithm.

8. The method for predicting the preventive risk index of downhole ventilation doors according to claim 1, characterized in that, The risk index prediction for the target damper includes: Obtain the parameter values ​​of the target damper at key monitoring points; Input the parameter values ​​into the risk prediction model, and output the predicted risk index of the target damper.

9. The method for predicting the preventive risk index of downhole ventilation doors according to claim 1, characterized in that, The determined reference damper set includes: Set a similarity interval for the historical risk index of the target damper. This similarity interval is centered on the historical risk index of the target damper and fluctuates by a fixed offset. Select other dampers whose historical risk indices fall within a similar range from all dampers to form a reference damper set.

10. The method for predicting the preventive risk index of downhole ventilation doors according to claim 1, characterized in that, The method further includes: Regularly update the operating status data, environmental condition data, and historical risk index of the damper, and retrain the risk prediction model.

Images