Upper corner gas monitoring reliability evaluation method and system based on working face associated sensor

Abstract

The invention relates to an upper corner gas monitoring reliability evaluation method and system based on a working face associated sensor, and belongs to the technical field of coal mine safety. The method comprises the following steps: firstly, synchronously acquiring historical data of gas concentration sensors at fixed positions of an intake airway, an upper corner and a return airway and production speed data of a production system, and preprocessing the acquired data and aligning time scales; then, building a gas concentration prediction model based on a random forest model by taking the upper corner gas concentration data as a benchmark; and finally, judging the reliability of the monitoring data by comparing the deviation between the real-time monitoring value and the predicted value of the upper corner gas concentration sensor, and if the deviation exceeds a preset threshold value, judging that the monitoring data is unreliable and triggering an alarm. The reliability of the upper corner gas monitoring data can be effectively evaluated, support is provided for accurate monitoring and early warning of a coal mine, and the safety management level of the coal mine is improved.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety technology and relates to a method and system for reliability assessment of gas monitoring in the upper corner based on working face associated sensors. Background Technology

[0002] With the continuous development of coal mining technology, coal mine safety has received increasing attention, and gas safety monitoring is a crucial aspect of coal mine safety. During coal mining, the upper corner is a high-risk area for gas accumulation in the working face, making the accuracy and reliability of gas concentration monitoring extremely important for coal mine safety. Currently, fixed gas sensors are commonly used in coal mine safety monitoring systems to monitor gas concentration in the upper corner in real time, enabling timely detection of gas exceedances and the implementation of appropriate measures.

[0003] In the prior art, CN109752504B discloses a decision-making auxiliary method for calibrating gas sensors in working faces. This method establishes a model using Gaussian process regression theory, analyzes and compares monitoring data before and after gas sensor calibration, and determines the effectiveness of the gas sensor calibration. CN102608286B proposes a method for real-time monitoring of abnormal gas monitoring values ​​in coal mines. This method establishes a confidence interval for gas concentration and determines whether the real-time monitoring value is within the confidence interval, thereby determining whether the monitoring value is abnormal. These methods improve the accuracy of gas monitoring to some extent, but none of them consider the impact of sensor position changes or human interference on monitoring reliability.

[0004] Regarding gas monitoring equipment in the upper corner, CN116838398B discloses an upper corner gas monitoring and emission facilitation device. This device achieves gas monitoring and emission in the upper corner through a combination of air inlet duct, air outlet duct, and ventilation duct, as well as uniformly distributed gas sensors. However, this device mainly focuses on gas monitoring and emission, without addressing the issue of assessing the reliability of the monitoring data.

[0005] CN110985129B proposes a method for identifying coal and gas outburst disasters in coal mine working faces. This method identifies coal and gas outbursts by analyzing the abnormal changes in gas concentration, wind speed, and wind direction data at different locations within the working face. Although this method utilizes the correlation of multi-sensor data, it primarily focuses on the identification of outburst disasters rather than the assessment of sensor monitoring reliability.

[0006] CN112177673B discloses a gas monitoring system for coal mining faces and a method for obtaining gas concentration distribution characteristics. By selecting multiple supports within the coal mining face and installing multiple gas sensors on each support, a three-dimensional gas concentration field is formed, allowing the determination of gas concentration fluctuation characteristics across the entire working face. While this system improves the comprehensiveness of gas monitoring, it does not solve the problem of the reliability of monitoring by individual key location sensors.

[0007] However, existing technologies have the following shortcomings: First, existing gas monitoring systems mainly focus on data acquisition and anomaly detection, lacking a mechanism for assessing the reliability of the monitoring data. Second, in actual coal mine production, improper operation by workers, such as moving sensors in the upper corner to other locations or covering sensors to avoid alarms, leads to distorted gas monitoring data, failing to accurately reflect the gas concentration in the upper corner and seriously threatening safe coal mine production. Third, existing technologies fail to fully utilize the data correlation between sensors at various locations on the working face, lacking a gas concentration estimation model for the upper corner based on multi-sensor data fusion, and thus cannot effectively verify the reliability of the gas sensor monitoring data in the upper corner.

[0008] Therefore, there is an urgent need for a method to assess the reliability of gas monitoring in the upper corner of the working face. This method involves analyzing the correlation between sensor data at various locations on the working face, establishing a gas concentration estimation model for the upper corner, and comparing the estimated values ​​with the monitored values ​​to determine the reliability of the monitoring data. This would allow for the timely detection of monitoring data anomalies caused by human interference, thereby ensuring safe production in coal mines. Summary of the Invention

[0009] In view of this, the purpose of this invention is to provide a method and system for reliability assessment of gas monitoring in the upper corner based on working face-related sensors.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] On the one hand, a reliability assessment method for gas monitoring in the upper corner based on working face-related sensors is proposed, characterized in that: the method includes:

[0012] S1. Synchronously acquire historical monitoring data of gas concentration from the working face intake gas concentration sensor A, upper corner gas concentration sensor B, and return air gas concentration sensor C from the coal mine safety monitoring system, and collect production speed data from the production system.

[0013] S2. The acquired data is preprocessed, and then the data from the gas concentration sensor B at the upper corner is used as a reference to align the data from sensor A, sensor C, and production speed on the event basis.

[0014] S3. Based on the data at each time point after alignment, using the gas concentration data and production speed data of sensors A and C as inputs and the gas concentration data of sensor B as output, establish a gas concentration prediction model at sensor B.

[0015] S4. Based on the established gas concentration prediction model, collect the current data of sensor A, sensor C and production speed, and predict the gas concentration value of sensor B at the current moment.

[0016] S5. Obtain the gas concentration monitoring value at the upper corner of the gas concentration sensor B from the coal mine safety monitoring system, calculate the deviation between the sensor monitoring value and the gas concentration prediction value. If the deviation value exceeds the preset threshold, the gas concentration monitoring value at the current moment is determined to be unreliable; if the deviation value does not exceed the preset threshold, the gas concentration detection value at the current moment is determined to be reliable.

[0017] Furthermore, for the sensor data at location A, the set of gas concentration data acquired is denoted as C. A ={C A,1 C A，2 ,…,C A,a}, where C A,i ∈C A ,i∈[1,a] represents the i-th gas concentration sampling data at location point A;

[0018] For the sensor data at location B, the set of gas concentration data acquired is denoted as C. B ={C B,1 C B，2 ,…,C B,b}, where C B,i ∈C B ,i∈[1,b] represents the i-th gas concentration sampling data at location point B;

[0019] For sensor data at location C, the set of acquired gas concentration data is denoted as C0. C ={C C,1 C C,2 ,…,C C,c}, where C C,i ∈C C ,i∈[1,c] represents the gas concentration sampling data of the i-th location point C;

[0020] The production speed data of the acquired production system is represented as P = {P1, P2, ..., P}. n}, where P i ,i∈[1,n] represents the i-th production speed data sampled.

[0021] Furthermore, median filtering or moving average was first used to remove peak noise from the gas concentration data; for the production speed data, outliers exceeding the physical range were removed.

[0022] Then, time alignment is performed on the data, the process of which is as follows:

[0023] Construct a distance matrix D between each data point and the baseline point B, where X∈{C} A C C}, then the distance matrix D is expressed as:

[0024] D X (i,j)=||X i |-|C B,j ||

[0025] Among them, X i This represents the i-th gas concentration sampling data at location point A or location point C; the subscript X represents the data set label for location point A or location point C.

[0026] Recursively calculate the cumulative distance and use dynamic programming to fill the cumulative distance matrix Γ:

[0027]

[0028] Backtracking the optimal path, tracing the optimal path K backwards from Γ(m,n). * ={(i k ,j k )}, which represents the optimal matching point pair between the gas concentration at location X and location B, where m and n are the data sequence lengths of location X and location B, respectively;

[0029] Generate the aligned sequence for each time point t of each data point at location B. j Find path K * The most recent time point of the data at the midpoint X The sequence corresponding to the points in time;

[0030] For the production speed data of the production system, the time point of the nearest location point B is directly matched.

[0031] Furthermore, for the aligned feature data, a random forest model is used as the base model to construct a gas concentration prediction model. The random forest model is composed of multiple decision tree-based learners. Each decision tree has a tree structure, including a root node, internal nodes, and leaf nodes. The root node is used to hold the initial training data. The internal nodes represent feature judgment conditions, and the data is divided into different child nodes according to the conditions. The leaf nodes are terminal nodes, and their values ​​are the average gas concentration of the sample at location B of the corresponding node.

[0032] An input layer is set up to receive historical data at various time points after alignment processing, specifically covering the gas concentration at location A, the gas concentration at location C, and the production speed data of the production system.

[0033] An output layer is set up to synthesize the prediction results of all decision trees and obtain the final gas concentration prediction value at location point B through an averaging method.

[0034] Furthermore, in the random forest model processing, the Bootstrap sampling method is used to randomly sample with replacement from the original training dataset to form training subsets for each decision tree.

[0035] When performing feature partitioning at each internal node, a subset of features is randomly selected from all input features. Based on the relationship between the features and the gas concentration at location point B, the optimal partitioning features and partitioning points are chosen using indicators such as the Gini index. The formula for calculating the Gini index is:

[0036]

[0037] Where D is the current node's dataset, K is the number of categories, and p k Let be the proportion of samples in dataset D belonging to the k-th class;

[0038] The node data is divided according to the selected optimal splitting features and splitting points until the stopping conditions are met. The stopping conditions include: the number of node samples is less than the preset minimum number of samples, the tree depth reaches the preset maximum depth, or the Gini index improvement is less than a specific threshold.

[0039] When the stopping condition is met, the leaf node is assigned the average gas concentration at location point B of the sample at that node.

[0040] Furthermore, during the training of the random forest model, the number of decision trees N, the maximum depth of each tree, the minimum number of samples per node, and the number of randomly selected features are pre-set. The k-fold cross-validation method is used to evaluate the model performance. The training dataset is divided into k subsets, and k-1 subsets are used for training in turn, while the remaining subsets are used for validation. The average performance index is taken as the evaluation result. Based on the cross-validation results, the model parameters are adjusted using methods such as grid search.

[0041] For a new input sample, each decision tree, based on its own structure and partitioning rules, traverses from the root node to the leaf node to obtain the predicted value of the gas concentration at location point B.

[0042] The predictions from all decision trees are integrated using an averaging method, as shown in the formula:

[0043]

[0044] Where N is the number of decision trees. Let be the predicted value of the nth decision tree. This represents the model's final predicted value.

[0045] Furthermore, the reliability of the upper corner gas concentration sensor at the current moment is determined based on the relationship between the predicted value, the monitored value, and the threshold k. Specifically, when... When the threshold is exceeded, the result is considered unreliable. k is determined based on the maximum error of the model training in the previous period. If the reliability results are monitored for several consecutive time points, and all results are unreliable, an alarm is triggered. If the result is unreliable only at a single time point, the alarm is ignored.

[0046] On the other hand, a reliability assessment system for gas monitoring in the upper corner based on working face-related sensors is also proposed, which includes a data acquisition unit, a data processor, and an alarm device.

[0047] The data acquisition unit periodically collects data from the corresponding sensors at various locations in the coal mine safety monitoring system, and simultaneously obtains the production speed from the production system and sends it to the data processor.

[0048] The data processor evaluates the reliability of the gas concentration sensor's monitoring value based on the aforementioned reliability evaluation method for gas monitoring in the upper corner based on working face associated sensors. If the evaluation result is unreliable, an alarm command is immediately sent to the alarm device.

[0049] Upon receiving an alarm command, the alarm device immediately activates an audible and visual alarm and simultaneously sends alarm information to relevant management personnel.

[0050] The beneficial effects of this invention are as follows:

[0051] This invention acquires gas concentration data from the intake airway, upper corner, and return airway of the coal mine safety monitoring system, and simultaneously acquires production speed data from the production system. Based on the gas concentration data and production speed data from multiple locations, a comprehensive and rich data system is constructed to predict the gas concentration in the upper corner. These data are not isolated but interconnected and mutually influential, collectively reflecting the complex environmental conditions underground in coal mines.

[0052] This invention further establishes a predictive model for gas concentration in the upper corner. This model is not a simple linear combination, but rather fully considers the complex relationships between various factors. By deeply analyzing the intrinsic connection between gas concentration changes in the intake and return airways and the gas concentration in the upper corner, as well as the influence mechanism of production speed on gas concentration, the model can more accurately predict the gas concentration in the upper corner.

[0053] In actual monitoring, when there is a significant deviation between the monitored data and the model's predicted values, it is possible to promptly detect monitoring anomalies caused by potential human interference or sensor malfunction. For example, if the sensor is improperly operated or malfunctions, the monitored gas concentration data will differ from the model's prediction based on multi-source data. Through this comparative analysis, the reliability of the gas concentration data in the upper corner can be determined in a timely manner, thereby accurately grasping the gas accumulation status of the working face and providing a guarantee for coal mine safety monitoring and disaster early warning.

[0054] This invention employs a random forest model for methane concentration estimation. Random forest, as an ensemble learning method, offers numerous advantages. It consists of multiple decision trees, each capable of independently analyzing and predicting data. The results of all decision trees are then integrated, significantly improving prediction accuracy. In the complex and variable environment of coal mines, methane concentration is influenced by the interaction of various factors, and traditional single models often struggle to accurately capture these complex relationships. However, the random forest model, with its powerful nonlinear fitting ability and capacity to handle high-dimensional data, is better suited to this complex environment. It can automatically learn features and patterns in the data, effectively handle noisy data and outliers, and exhibits high robustness.

[0055] By employing a random forest model for methane concentration estimation, this invention provides a strong guarantee for safe coal mine production. Accurate methane concentration prediction can help coal mining enterprises formulate reasonable ventilation strategies and production plans in advance, avoiding safety accidents caused by methane accumulation. Simultaneously, timely detection and intervention of monitoring anomalies can reduce misjudgments and malfunctions caused by inaccurate monitoring data, thereby improving the safety of coal mine production.

[0056] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0057] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0058] Figure 1 This is a flowchart illustrating the reliability assessment method for upper corner gas monitoring based on working face-associated sensors according to an embodiment of the present invention.

[0059] Figure 2 This is a schematic diagram of the sensor deployment location according to an embodiment of the present invention;

[0060] Figure 3 This is a schematic diagram of the structure of the upper corner gas monitoring reliability assessment system based on working face associated sensors according to an embodiment of the present invention. Detailed Implementation

[0061] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0062] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0063] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0064] Please see Figures 1-3 This invention relates to a reliability assessment method and system for gas monitoring in the upper corner of the working face based on a working face-correlated sensor.

[0065] Example 1

[0066] This embodiment provides a reliability assessment method for upper corner gas monitoring based on working face-associated sensors. This method collects multi-source data by associating fixed-location gas sensors on the coal mine working face, establishes a gas concentration prediction model, and thus assesses the reliability of upper corner gas monitoring data. Figure 1 As shown, the specific implementation steps are as follows:

[0067] Step 1: Data Acquisition

[0068] like Figure 2 As shown, the processor synchronously acquires historical gas concentration monitoring data from the gas concentration sensors A (intake airway), B (upper corner), and C (return airway) at the working face from the coal mine safety monitoring system, and also collects production speed data from the production system. The processor synchronously acquires data from the gas concentration sensors at locations A (intake airway), B (upper corner), and C (return airway) at the working face from the coal mine safety monitoring system; simultaneously, the processor also connects to the production system and acquires the production speed according to a preset transmission cycle.

[0069] These sensors all employ high-precision gas detection equipment, with a detection accuracy of no less than 0.01% and a response time of no more than 10 seconds.

[0070] For the sensor data at location A, the set of gas concentration data acquired is denoted as C. A ={C A,1 C A，2 ,…,C A,a}, where C A,i ∈C A ,i∈[1,a] represents the i-th gas concentration sampling data at location point A;

[0071] For the sensor data at location B, the set of gas concentration data acquired is denoted as C. B ={C B,1 C B，2 ,…,C B,b}, where C B,i ∈C B ,i∈[1,b] represents the i-th gas concentration sampling data at location point B;

[0072] For sensor data at location C, the set of acquired gas concentration data is denoted as C0. C ={C C,1 C C,2 ,…,C C,c}, where C C,i ∈C C ,i∈[1,c] represents the gas concentration sampling data of the i-th location point C;

[0073] The production speed data of the acquired production system is represented as P = {P1, P2, ..., P}. n}, where P i ,i∈[1,n] represents the i-th production speed data sampled.

[0074] Historical monitoring data from the inlet gas concentration sensor, the upper corner gas concentration sensor, and the outlet gas concentration sensor are acquired simultaneously based on time series data. Additionally, production speed data is collected from the production system at a sampling frequency of once every 5 seconds. In a preferred embodiment, the data acquisition time span is no less than 30 days to ensure the comprehensiveness and representativeness of the data, covering production conditions under different operating circumstances.

[0075] Step 2: Data Preprocessing

[0076] The collected raw data underwent preprocessing, including outlier detection and removal, data smoothing, and normalization. Outlier detection employed the 3σ principle, treating data exceeding the mean ± 3 standard deviations as outliers and removing them. Data smoothing used a moving average method with a window size of 5 sampling points. Normalization mapped all data types to the [0,1] interval, facilitating subsequent model building.

[0077] Step 3: Data Time Alignment

[0078] The process of aligning data by time is as follows:

[0079] Construct a distance matrix D between each data point and the baseline point B, where X∈{C} A C C}, then the distance matrix D is expressed as:

[0080] D X (i,j)=||X i |-|C B,j ||

[0081] Among them, X i This represents the i-th gas concentration sampling data at location point A or location point C; the subscript X represents the data set label for location point A or location point C.

[0082] Recursively calculate the cumulative distance and use dynamic programming to fill the cumulative distance matrix Γ:

[0083]

[0084] Backtracking the optimal path, tracing the optimal path K backwards from Γ(m,n). * ={(i k ,j k )}, which represents the optimal matching point pair between the gas concentration at location X and location B, where m and n are the data sequence lengths of location X and location B, respectively;

[0085] Generate the aligned sequence for each time point t of each data point at location B. j Find path K * The most recent time point of the data at the midpoint X The sequence corresponding to the points in time;

[0086] For the production speed data of the production system, the time point of the nearest location point B is directly matched.

[0087] Step 4: Establishing a gas concentration estimation model

[0088] Based on the aligned historical data at each time point, a gas concentration estimation model for location B is established, using the gas concentration data at location A and location C, the production speed of the production system, and the gas concentration at location B as the output.

[0089] For the aligned feature data, a random forest model is used as the base model to construct a gas concentration prediction model. The random forest model is composed of multiple decision tree-based learners. Each decision tree has a tree structure, including a root node, internal nodes, and leaf nodes. The root node is used to hold the initial training data. The internal nodes represent feature judgment conditions, and the data is divided into different child nodes according to the conditions. The leaf nodes are terminal nodes, and their values ​​are the average gas concentration of the sample at location B of the corresponding node.

[0090] An input layer is set up to receive historical data at various time points after alignment processing, specifically covering the gas concentration at location A, the gas concentration at location C, and the production speed data of the production system.

[0091] An output layer is set up to synthesize the prediction results of all decision trees and obtain the final gas concentration prediction value at location point B through an averaging method.

[0092] Furthermore, in the random forest model processing, the Bootstrap sampling method is used to randomly sample with replacement from the original training dataset to form training subsets for each decision tree.

[0093] When performing feature partitioning at each internal node, a subset of features is randomly selected from all input features. Based on the relationship between the features and the gas concentration at location point B, the optimal partitioning features and partitioning points are chosen using indicators such as the Gini index. The formula for calculating the Gini index is:

[0094]

[0095] Where D is the current node's dataset, K is the number of categories, and p k Let be the proportion of samples in dataset D belonging to the k-th class;

[0096] The node data is divided according to the selected optimal splitting features and splitting points until the stopping conditions are met. The stopping conditions include: the number of node samples is less than the preset minimum number of samples, the tree depth reaches the preset maximum depth, or the Gini index improvement is less than a specific threshold.

[0097] When the stopping condition is met, the leaf node is assigned the average gas concentration at location point B of the sample at that node.

[0098] During the training of the random forest model, the number of decision trees N, the maximum depth of each tree, the minimum number of samples per node, and the number of randomly selected features are pre-set. The k-fold cross-validation method is used to evaluate the model performance. The training dataset is divided into k subsets, and k-1 subsets are used for training in turn, while the remaining subsets are used for validation. The average performance index is taken as the evaluation result. Based on the cross-validation results, the model parameters are adjusted by methods such as grid search.

[0099] The model was built using the random forest algorithm, which has good nonlinear fitting ability and anti-interference ability. The main parameters of the model were set as follows: 100 decision trees, a maximum depth of 10 for each tree, a minimum number of leaf node samples of 5, and the Gini coefficient as the feature selection criterion.

[0100] In a preferred embodiment, the model is built using the support vector regression algorithm, the kernel function is selected as the RBF kernel, the penalty factor C is set to 10, the epsilon is set to 0.01, and the gamma parameter is selected automatically.

[0101] In another preferred embodiment, the model is built using a deep neural network with a network structure of 3 fully connected layers, each with 64, 32, and 16 neurons, respectively. The activation function is ReLU, the optimizer is Adam, the learning rate is set to 0.001, and the training epochs are 200.

[0102] Step 5: Gas Concentration Prediction

[0103] Based on the established gas concentration prediction model, gas concentration data at locations A and C at the current moment, as well as the production speed of the production system, are collected to predict the gas concentration at location B at the current moment. The model input data undergoes the same preprocessing procedures as the training data, including outlier detection, data smoothing, and normalization, to ensure the consistency of the input data.

[0104] For a new input sample, each decision tree, based on its own structure and partitioning rules, traverses from the root node to the leaf node to obtain the predicted value of the gas concentration at location point B.

[0105] The predictions from all decision trees are integrated using an averaging method, as shown in the formula:

[0106]

[0107] Where N is the number of decision trees. Let be the predicted value of the nth decision tree. The model's final prediction.

[0108] Step Six: Reliability Assessment

[0109] The gas concentration monitoring value from the upper corner gas concentration sensor at location B is obtained from the coal mine safety monitoring system, and the deviation between the monitored gas concentration value and the predicted gas concentration value is calculated. If the deviation value exceeds a preset threshold, the current gas concentration monitoring value is considered unreliable; if the deviation value does not exceed the preset threshold, the current gas concentration monitoring value is considered reliable.

[0110] The preset threshold is determined by analyzing historical data to calculate the mean absolute error (MAE) of the model on the validation set, and then setting the threshold to twice the MAE. In practical applications, the threshold is typically set between 0.05% and 0.1%, and can be adjusted according to specific working conditions.

[0111] In a preferred embodiment, the reliability assessment result is divided into three levels: when the deviation value is less than 50% of the threshold, it is judged as "high reliability"; when the deviation value is between 50% and 100% of the threshold, it is judged as "medium reliability"; and when the deviation value is greater than the threshold, it is judged as "unreliable".

[0112] In another preferred embodiment, the system will also record the number of consecutive unreliable judgments. When there are three or more consecutive unreliable judgments, the system will issue an alarm signal to prompt relevant personnel to check whether the gas concentration sensor in the upper corner is faulty or abnormal.

[0113] This method establishes a predictive model for gas concentration in the upper corner of the coal mine through multi-source data fusion and machine learning techniques, enabling real-time assessment of the reliability of gas monitoring data and effectively improving the safety assurance capability of coal mine production. In practical applications, this method can detect anomalies in monitoring data caused by sensor malfunctions, calibration deviations, or improper installation locations, providing a basis for the maintenance and optimization of the gas monitoring system.

[0114] Example 2

[0115] This embodiment provides a reliability assessment system for gas monitoring in the upper corner based on working face-related sensors, such as... Figure 3 As shown, it includes corresponding gas concentration sensors installed at the intake airway A, upper corner B, and return airway C in the return airway, and also includes a data acquisition unit, a data processor, and an alarm device.

[0116] The data acquisition unit periodically collects data from the corresponding sensors at various locations in the coal mine safety monitoring system, and simultaneously obtains the production speed from the production system and sends it to the data processor.

[0117] The data processor evaluates the reliability of the gas concentration sensor's monitoring value based on the aforementioned reliability evaluation method for gas monitoring in the upper corner based on working face associated sensors. If the evaluation result is unreliable, an alarm command is immediately sent to the alarm device.

[0118] Upon receiving an alarm command, the alarm device immediately activates an audible and visual alarm and simultaneously sends alarm information to relevant management personnel.

[0119] The data acquisition unit is configured to periodically collect data from various sensors every 5 seconds. It connects to the coal mine safety monitoring system via a wired network, periodically collecting data from corresponding sensors at each location to ensure stable and real-time data transmission. Simultaneously, the data acquisition unit obtains production speed data from the production system's monitoring host. This production speed data reflects the production intensity at the working face. All collected data is packaged and sent to the data processor via industrial Ethernet.

[0120] After receiving the data, the data processor first performs data preprocessing. For gas concentration data, a median filtering algorithm is used to remove peak noise. For example, for gas concentration data in intake roadway A, the median of five consecutive sampling points is taken as the valid data at that moment. For production speed data, outliers that clearly exceed the physical range are removed. For example, the speed of the coal mining machine should not exceed 1.2 times its maximum design speed. If the collected data exceeds this range, it is considered an outlier and is removed.

[0121] Next, time alignment is performed, using the data time point at the upper corner B as the baseline. The Dynamic Time Warping (DTW) algorithm is employed to align the data from intake airway A and return airway C, as well as the production speed data of the production system, in time. By calculating the time deviation between data points at different locations and using linear interpolation, all data are made consistent along the time axis, providing an accurate data foundation for subsequent reliability assessments.

[0122] The data processor, based on the aforementioned reliability assessment method for upper corner gas monitoring using working face-related sensors, utilizes a random forest model to predict the upper corner gas concentration. The model is trained using historical data, with inputs including the gas concentrations at locations A and B and the production speed of the production system, and the output being the predicted gas concentration at location B.

[0123] In the real-time evaluation phase, the aligned data of the current moment is input into the trained model to obtain the predicted gas concentration value. This value is then compared with the monitored value from the gas concentration sensor at the upper corner B, and the deviation between the two is calculated. The reliability of the upper corner gas concentration sensor at the current moment is determined based on the relationship between the predicted and monitored values ​​and a threshold k. When the threshold is exceeded, the model is deemed unreliable. k is determined based on the maximum error during the initial model training.

[0124] When the data processor determines the data to be unreliable, it immediately sends an alarm command to the alarm device. Upon receiving the command, the alarm device immediately activates an audible and visual alarm, with bright flashes and a loud siren sounding in the central control room underground to alert the on-duty personnel. Simultaneously, the alarm device sends alarm information, including the time of the unreliability and the deviation between the predicted and monitored values, to relevant management personnel via SMS and the underground communication system.

[0125] Upon receiving the alarm, relevant management personnel acted swiftly. Firstly, maintenance personnel were dispatched to location B in the upper corner to inspect and repair the gas concentration sensor, investigating for any malfunctions or human interference. Secondly, based on the assessment results and actual conditions, the operating parameters of the production system were adjusted, such as reducing the speed of the coal mining machine, to minimize the risk of gas outbursts. These measures ensured the reliability and safety of coal mine gas monitoring, guaranteeing normal production at the 301 longwall mining face.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A working face correlation sensor-based reliability evaluation method for gas monitoring in an upper corner, characterized in that: The method comprises: S1, synchronously acquiring gas concentration historical monitoring data of an air inlet gas concentration sensor A, an upper corner gas concentration sensor B and an air return gas concentration sensor C from a coal mine safety monitoring system, and collecting production speed data from a production system; S2, preprocessing the acquired data, and taking the data at the upper corner gas concentration sensor B as a reference, aligning the data of the sensors A and C and the production speed at events; S3, based on the data at each time point after alignment, taking the gas concentration data of the sensors A and C and the production speed data as inputs, and taking the gas concentration data of the sensor B as an output, establishing a gas concentration prediction model at the sensor B; S4, according to the established gas concentration prediction model, collecting the data of the sensor A, the sensor C and the production speed at the current time, and predicting the gas concentration value of the sensor B at the current time; S5, acquiring the upper corner gas concentration monitoring value of the upper corner gas concentration sensor B from the coal mine safety monitoring system, calculating the deviation value between the sensor monitoring value and the gas concentration prediction value, if the deviation value exceeds a preset threshold, judging that the gas concentration monitoring value at the current time is unreliable, and if the deviation value does not exceed the preset threshold, judging that the gas concentration monitoring value at the current time is reliable.

2. The working face correlation sensor-based reliability evaluation method for upper corner gas monitoring according to claim 1, characterized in that: For the sensor data of the position point A, the obtained gas concentration data set is denoted as C A ={C A,1 ,C A，2 ,…,C A,a} , wherein C A,i ∈C A ,i∈[1,a] represents the i th gas concentration sampling data of the position point A; For the sensor data of the position point B, the obtained gas concentration data set is denoted as C B ={C B,1 1,C B，2 2,…,C B,b b}, wherein C B,i i∈C B ,i∈[1,b] represents the i th gas concentration sampling data of the position point B; For the sensor data of the position point C, the obtained gas concentration data set is denoted as C C = {C C,1 1, C C,2 2, …, C C,c c}, wherein C C,i i∈C C ,i∈[1,c] represents the i-th gas concentration sampling data of the position point C; For the acquired production speed data of the production system, it is represented as P = {P1, P2, … Pn}, where P n ,i∈[1,n] represents the i-th production speed data of the sampling. i ,i∈[1,n] represents the i-th production speed data of the sampling.

3. The working face correlation sensor-based reliability evaluation method for upper corner gas monitoring according to claim 2, characterized in that: First, the median filter or moving average is used to remove the peak noise of the gas concentration data; for the production speed data, the abnormal values exceeding the physical range are removed; Then, the data is time-aligned, and the process is as follows: A distance matrix D is constructed for each data point B, where X e {C A ,C C} and D is given by: D X (i,j) = ||X i |-|C B,j || wherein X i represents the i-th gas concentration sampling data of the position point A or the position point C; the subscript X represents the data set label of the position point A or the position point C; The cumulative distance is recursively calculated, and the cumulative distance matrix Γ is filled using dynamic programming: Backtracking the optimal path, tracing the optimal path K from Γ(m, n) in reverse * = {(i k ,j k )}, which represents the gas concentration at the position point X and the optimal matching point pair of the position point B, m, n are the data sequence lengths of the position point X and the position point B respectively; generating an aligned sequence for each time point t of each data in position point B j finding the nearest time point of data at position point X in path K * in path K sequences corresponding to time points For the production speed data of the production system, the time point of the nearest position point B is directly matched.

4. The working face correlation sensor-based reliability evaluation method for upper corner gas monitoring according to claim 1, characterized in that: For the aligned feature data, a random forest model is used as a basic model to construct a gas concentration prediction model, wherein the random forest model is composed of multiple decision tree base learners, each decision tree has a tree structure, including a root node, an internal node and a leaf node; the root node is used to accommodate the initial training data; the internal node represents a feature judgment condition, and the data is divided into different sub-nodes according to the condition; the leaf node is a terminal node, and its value is the average gas concentration of the position point B of the samples contained in the node; The input layer is set to receive the historical data at each time point after alignment, specifically including the gas concentration of the position point A, the gas concentration of the position point C and the production speed data of the production system; The output layer is set to integrate the prediction results of all decision trees, and the final position point B gas concentration prediction value is obtained by an average method.

5. The working face correlation sensor-based reliability evaluation method for upper corner gas monitoring according to claim 4, characterized in that: In the model processing process of the random forest, the Bootstrap sampling method is used to randomly sample samples from the original training data set with replacement to form the training subsets of each decision tree; When dividing features at each internal node, a part of features is randomly selected from all input features, and the best division feature and division point are selected based on the relationship between the feature and the gas concentration of the position point B according to indicators such as Gini index; the Gini index calculation formula is: where D is the current node dataset, K is the number of classes, p k is the proportion of samples in dataset D that belong to the kth class. According to the selected optimal partition feature and partition point, the node data is partitioned until the stop condition is met; the stop condition includes: the number of node samples is less than the preset minimum sample number, the tree depth reaches the preset maximum depth, or the Gini index improvement is less than a specific threshold value; When the stop condition is reached, the leaf node is assigned as the mean value of the location point B gas concentration of the node samples.

6. The working face correlation sensor-based reliability evaluation method for upper corner gas monitoring according to claim 5, characterized in that: In the process of training the random forest model, the number of decision trees N, the maximum depth of each tree, the minimum sample number of node partition, and the number of randomly selected features are preset; the k-fold cross-validation method is used to evaluate the model performance, the training data set is divided into k subsets, k-1 subsets are used for training and the remaining subset is used for verification, and the average performance index is taken as the evaluation result; according to the cross-validation result, the model parameters are adjusted by grid search method; For new input samples, each decision tree traverses from the root node to the leaf node according to its own structure and partition rule, and obtains the predicted value of the location point B gas concentration; The predicted values of all decision trees are integrated by using the average method, and the formula is: where N is the number of decision trees, is the prediction value of the nth decision tree, is the final prediction value of the model.

7. The working face correlation sensor-based reliability evaluation method for upper corner gas monitoring according to claim 6, characterized in that: According to the relationship between the predicted value and the monitoring value and the threshold value k, the reliability of the upper corner gas concentration sensor at the current time is determined, wherein when , that is, exceeding the critical value, it is determined to be unreliable, and k is determined according to the maximum error of the previous model training; if the reliability results of several time points are continuously monitored, and if they are all unreliable, an alarm is triggered; if only the result of a single time point is unreliable, the alarm is ignored.

8. A working face correlation sensor-based reliability evaluation system for gas monitoring in an upper corner, characterized in that: It includes a data collector, a data processor, and an alarm device, wherein, The data collector periodically collects corresponding position sensor data from the coal mine safety monitoring system and synchronously obtains the production speed from the production system, and sends it to the data processor; The data processor evaluates the reliability of the monitoring value of the upper corner gas concentration sensor according to the working face related sensor based upper corner gas monitoring reliability evaluation method of any one of claims 1-7, and if the evaluation result is unreliable, an alarm instruction is immediately sent to the alarm device; After receiving the alarm instruction, the alarm device immediately performs sound and light alarm, and synchronously sends alarm information to the relevant management personnel.

Images