A coal and gas outburst early warning method based on field real-time data driving

Abstract

This invention belongs to the field of coal mine safety production warning technology and discloses a coal and gas outburst early warning method based on real-time on-site data. By continuously acquiring real-time environmental data on gas and wind speed through gas concentration sensors and wind speed sensors installed at the tunneling face, the gas emission rate is calculated. In-depth analysis of the gas emission rate yields early warning indicators for coal and gas outbursts, combining dynamic changes in gas concentration with reverse identification based on gas emission rate prediction and evaluation indicators. Moving average, deviation rate, dispersion rate, volatility, root mean square error, and mean absolute percentage error early warning indicators are established. The weights of each indicator are determined using the analytic hierarchy process (AHP), a fuzzy comprehensive early warning model is constructed, and a comprehensive index of the differences among the indicators is calculated for coal and gas outburst early warning. This invention solves the problem of existing methods' difficulty in real-time data acquisition, which hinders effective on-site decision-making.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety production warning technology, and in particular relates to a coal and gas outburst early warning method based on real-time on-site data. Background Technology

[0002] Currently, as coal and gas outburst accidents continue to occur, early warning models for coal and gas outbursts mainly select indicators such as gas content, gas pressure, gas emission, drill cuttings method, and electromagnetic radiation method, which can theoretically achieve good prediction results.

[0003] In coal and gas outburst early warning models, the gas content and pressure, key controlling factors, cannot be obtained in real time. Existing mechanistic models, employing methods such as drill cuttings analysis and electromagnetic radiation analysis, are susceptible to human intervention and the underground geological environment, resulting in predictive results with a certain degree of lag and significant errors. These factors mean that while these models can theoretically achieve good predictive results, their practical application is often limited by the real-time nature of data acquisition, preventing the predictions from effectively guiding on-site decision-making. Furthermore, relying solely on statistical methods or artificial intelligence algorithms to analyze and predict gas concentrations also has its limitations and cannot achieve satisfactory results.

[0004] Based on the above analysis, the problems and shortcomings of existing technologies are as follows: Most existing technologies attempt to analyze the pre-disaster evolution patterns based on the coal and gas outburst mechanism, and then assess the risk level of collectable data sequences through these evolution patterns to achieve early warning. Risk assessment based on evolution patterns requires the collection of a wide variety of data, most of which is difficult to obtain in real time. Furthermore, it is easily affected by human operation and the underground geological environment, resulting in a certain degree of lag in the prediction results and poor accuracy, making it difficult to guarantee the effectiveness of early warnings.

[0005] The difficulty in solving the above problems and shortcomings lies in the need to develop a large number of online sensors that can acquire parameters related to the evolution mechanism in order to ensure the accuracy and effectiveness of the predictions made by the above technical approach.

[0006] The significance of addressing the above problems and shortcomings is as follows: Since gas emission volume best characterizes the differences between different stages of the entire coal and gas outburst process, the variation pattern of gas emission volume during normal coal mine tunneling implicitly contains the variation pattern of gas emission volume during the outburst incubation stage. If abnormal changes that contradict this pattern can be identified, it can be determined that the coal mine tunneling face is in the formation and development stage of an outburst during that period. This invention uses statistical indicators combined with artificial intelligence model indicators to mine abnormal changes in gas emission volume, thereby determining the degree of coal and gas outburst risk. Gas emission volume can be obtained and calculated in real time using existing online gas concentration and wind speed sensors, without human interference, thus solving the problems of poor accuracy and effectiveness of traditional methods. Summary of the Invention

[0007] To overcome the problems existing in related technologies, the present invention discloses an embodiment of a coal and gas outburst early warning method based on real-time on-site data. Specifically, it relates to a coal and gas outburst early warning method based on the abnormal characteristics of gas emission time-series data.

[0008] The technical solution is as follows: a coal and gas outburst early warning method based on real-time on-site data.

[0009] By continuously acquiring real-time data on gas and wind speed from gas concentration and wind speed sensors installed at the tunneling face, the gas emission rate is calculated. In-depth analysis of the gas emission rate yields early warning indicators for coal and gas outbursts, combining dynamic changes in gas concentration with reverse identification based on gas emission rate prediction and evaluation indicators. Early warning indicators are established using moving average, deviation rate, dispersion rate, volatility, root mean square error (RMSE), and mean absolute percentage error (MAPE). The weights of each indicator are determined using the Analytic Hierarchy Process (AHP), a fuzzy comprehensive early warning model is constructed, and a comprehensive index of the differences between each indicator is calculated for early warning of coal and gas outbursts.

[0010] In one embodiment, the coal and gas outburst early warning method based on real-time on-site data specifically includes:

[0011] Step 1: By continuously collecting and monitoring the gas concentration and wind speed data of the return airway at the tunneling face, and combining the cross-sectional area of ​​the return airway, the gas emission rate is calculated.

[0012] Step two involves performing a five-point cubic smoothing process (using the least squares method to approximate the sampling points) on the obtained gas emission data to remove interference. The five-point cubic smoothing method is a time series smoothing technique that uses a cubic least squares polynomial to smooth the time series, specifically including:

[0013] There are n+1 uniform sampling points Y1, Y2, ..., Y n-1 Yn, with a sampling interval of h, undergo transformation.

[0014] μ i =(t i -t0) / h, i=0, 1, 2..., n;

[0015] Then the original nodes become μ0 = 0, μ1 = 1, μ2 = 2, ..., μ n-1 =n-1,μ n =n;

[0016] The data Y is fitted using an m-th degree polynomial. i Let the fitted polynomial be:

[0017] Y (μ) =a0+μa1+μ 2 a2+…+μ m-1 a m-1 +μ m a m

[0018] The undetermined coefficients a in the fitted polynomial i The variance sum is determined by the least squares method as follows:

[0019]

[0020] If F(a0, a1, a2, ..., a...) m The minimum value for a is 1. j (j = 0, 1, 2, ..., m) Find the partial derivative, we have

[0021]

[0022] After sorting, we get:

[0023]

[0024] Substituting n=4 and m=3 into the formula Solve for a j j = 0, 1, 2, 3, then a j Substitution Y (μ) =a0+μa1+μ 2 a2+…+μ m-1 a m-1 +μ m a m ;

[0025] Let μ n =1, 2, 3, 4, The formula for five-point cubic smoothing is... For Y i The smoothed value;

[0026] Step 3: Use statistical methods to extract the moving average, deviation rate, dispersion rate, and volatility of the gas emission change value as characteristic indicators;

[0027] Step four: Collect the methane sensor and wind speed sensor data from the T2 working face during two months of normal coal mine tunneling production, and then use formula Q... s =60Cv s S calculates the time-series data of gas emission, where Q s The amount of gas emitted from the tunnel, m 3 / min; C is the gas concentration in the tunnel, %; v s Wind speed in the tunnel, m / s;

[0028] S is the cross-sectional area of ​​the roadway at the measurement point, in meters. 2 The calculated gas emission time-series data is input into a Bi-directional Long Short-Term Memory (BiLSTM) recurrent neural network model for training and optimization to minimize the model training error and establish a data model under normal production conditions. This data model is the Bi-directional Long Short-Term Memory (BiLSTM) recurrent neural network model with the smallest error. The parameters of the trained Bi-directional Long Short-Term Memory (BiLSTM) recurrent neural network model with the smallest error are set as follows:

[0029] The batch size of data in one batch is 32.

[0030] Learning rate = 1e -4 ;

[0031] The data was executed epoch 20 times;

[0032] The data splitting ratio (split_data_ratio) is 0.833.

[0033] The number of steps is lag_num = 16;

[0034] The dimension of the input value, input_dim_size, is 16;

[0035] The predicted output value dimension output_dim_size = 1;

[0036] Hidden layer dimension hidden_dim_size = 64;

[0037] The number of LSTM layers is lstm_layer_hums = 1;

[0038] Step 5: Input the gas emission data obtained from real-time collection and calculation into the data model described in Step 4 as a test dataset to obtain the gas emission prediction value. Calculate the loss value using the difference between the prediction value and the actual value. Use the root mean square error (RMSE) and mean absolute percentage error (MAPE) of the loss value as early warning indicators for coal and gas outbursts.

[0039] Step 6: Based on the Analytic Hierarchy Process (AHP), construct the judgment matrix A using the 1-9 scaling method, and calculate the largest eigenvalue λ.max The consistency of the judgment matrix is ​​checked using the index weight vector W to determine the weights of each index. Based on the principle of determining the membership function, the membership function of each index is determined through a large amount of experimental data. Using the fuzzy comprehensive evaluation model, a fuzzy comprehensive evaluation set is established, and the calculated evaluation set value represents the comprehensive index of the differences of each index.

[0040] Step 7: Analyze the gas emission during normal coal mine production using the comprehensive index of differences among various indicators. Multiply the maximum calculated result by a safety factor (1.2–1.5) to determine the difference threshold. Use this threshold to identify the calculated comprehensive index sequence of differences among various indicators and determine the degree of coal and gas outburst risk (those greater than or equal to the threshold are considered high risk).

[0041] In one embodiment, step three, which uses statistical methods to extract the moving average, deviation rate, dispersion rate, and volatility of the gas emission rate as characteristic indicators, specifically includes:

[0042] 1) Extract the moving average of the gas emission rate to obtain the trend of gas concentration time series data. The formula for calculating the average value of gas concentration data is:

[0043]

[0044] In the formula, C represents the average of n gas concentration data points; n is the number of records in the sequence within a certain time period; i Let be the gas concentration of the i-th record in the sequence;

[0045] 2) Extract the deviation rate of the gas emission rate change value to obtain the amount by which the real-time time series data of gas concentration deviates from the moving average of gas concentration for that time period; the deviation rate is calculated using the following formula:

[0046]

[0047] In the formula, C t This represents the gas concentration at time t; The average value of n gas concentration data;

[0048] 3) Extract the dispersion rate of the gas emission rate change value to obtain the degree of dispersion of the gas concentration signal sequence. The dispersion rate is calculated using the following formula:

[0049]

[0050] In the formula: C t The m represents the gas concentration at time t; m is the number of sequence records within a certain time period.

[0051] 4) Extract the volatility of gas emission changes to obtain the actual situation of gas concentration changes; volatility extraction includes calculation of amplitude change rate and frequency change rate.

[0052] Formula for calculating the rate of change of amplitude R OSC for:

[0053] R OSC (n)=[(C Tmax -C Tmin ) / C Tmax (4)

[0054] In the formula: R osc (n) represents the rate of change of amplitude over a time interval n; C Tmax C Tmin These represent the maximum and minimum gas concentrations within the time interval;

[0055] Formula for calculating frequency change rate R OFC for:

[0056] R OFC =f(C (t-n)-t (5)

[0057] In the formula: C (t-n)-t Let f be the gas concentration from time tn to time t; f() represents the number of times the change occurs.

[0058] In one embodiment, the bidirectional long short-term memory recurrent neural network model in step four includes:

[0059] At each time step, eight weights are used in a cycle: weights input to the forward and backward hidden layers (W1, W4); weights of the forward hidden layer (W2, W3); weights of the backward hidden layer (W6, W7); weights from the forward and backward hidden layers to the output layer (W5, W8); and weights of the forward hidden layer h`. t Backward hidden layer h t and output layer o t The calculation formulas are as follows:

[0060] h` t =f(w1x t +w2C t-1 +w3h t-1 (6)

[0061] h t =f(w4x t +w6C t-1 +w7h t-1 (7)

[0062] o t =g(w5h) t +w8ht (8).

[0063] In one embodiment, the formula for calculating the root mean square error in step five is:

[0064]

[0065] The formula for calculating the mean absolute percentage error is:

[0066]

[0067] In the formula: C i Indicates the actual gas concentration; This indicates the predicted gas concentration; n represents the number of samples.

[0068] In one embodiment, step six, determining the weights of each indicator, specifically includes: constructing a judgment matrix A using the 1-9 scaling method and calculating the largest eigenvalue λ. max And the indicator weight vector W, the consistency test of the judgment matrix; the expression of the consistency test judgment matrix is:

[0069]

[0070] In the formula: C1 is the consistency ratio; n is the number of comparison indicators; R1 is the average random consistency index. When C1 < 0.1, the constructed judgment matrix meets the requirements.

[0071] In one embodiment, step six, determining the membership function, includes:

[0072] Moving average: Utilizes the real-time moving average of gas concentration and the average gas concentration of the previous period. Divide by the ratio;

[0073] Deviation rate and dispersion rate: The value range is divided by comparing the real-time deviation rate and dispersion rate with the average value of the calculation period;

[0074] Volatility: The range of values ​​is divided by comparing the real-time volatility with the average value of the calculation period, and the larger value between the amplitude change rate and the frequency change rate is selected;

[0075] Root mean square error and mean absolute percentage error: The range of values ​​is divided by the ratio of the evaluation value of the gas concentration sequence prediction result before the real-time accident to the evaluation average value of the gas concentration sequence prediction result during normal production.

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082] In one embodiment, step six, establishing the fuzzy comprehensive evaluation model, includes:

[0083] a. Establishing a factor set

[0084] The factor set U is represented as U = (u1, u2, ..., u3) n ); where element u i Let i represent the influencing factors, where i = 1, 2, ..., n;

[0085] b. Establish an evaluation set

[0086] The evaluation set is represented as V = (v1, v2, ... v) m ), where element v j Let V represent the evaluation result, where j = 1, 2, ..., m; when V = {an accident occurred, no accident occurred}, V = {I, II};

[0087] c. Establish weight set

[0088] The elements in the factor set U have different levels of importance in the evaluation, and each element u... i Different weight sets W are given according to their importance. W is a fuzzy subset of the factor set U, denoted as W = (w1, w2, ... w...). n When the weight set is W = (w1, w2), w1 + w2 = 1.

[0089] d. Single-factor fuzzy evaluation

[0090] The evaluation object is the i-th factor u in the factor set. i To make a judgment, evaluate the j-th element v in the evaluation set. j The degree of membership is r ij Then for the i-th factor u i The evaluation result is represented as R. i ={r i1 r i2 , ...r im}, R i Represents a single-factor evaluation set;

[0091] e. Fuzzy Comprehensive Evaluation

[0092] Construct a multi-factor comprehensive evaluation matrix R, R = {R1, R2}, from the single-factor evaluation set. TUsing a weighted average model and following the multiplication operation of fuzzy matrices, the fuzzy comprehensive evaluation set B is obtained as follows:

[0093] B = WR = {b1, b2, ..., b} m} (18)

[0094] In the formula: b i Let be a fuzzy comprehensive evaluation index, where i = 1, 2, ..., m, representing the degree of membership of the evaluation object to the i-th element in the evaluation set V, considering all influencing factors.

[0095] In one embodiment, step seven uses the value of the fuzzy comprehensive evaluation set B to represent the comprehensive index of differences among various indicators. The root mean square error and mean absolute percentage error of the evaluation indicators are predicted by the model, and coal and gas outburst evaluation and early warning are carried out in combination with the dynamic change characteristics of gas concentration. During normal coal mine production, the comprehensive index of differences among various indicators fluctuates within the normal range. When the index reaches the maximum value within the normal range, an orange warning is issued. When the comprehensive index of differences among various indicators approaches 1, it indicates that the possibility of coal and gas outburst is greater, and a red warning is issued.

[0096] Combining all the above technical solutions, the advantages and positive effects of this invention are as follows: Since gas emission volume best characterizes the differences between different stages of the entire coal and gas outburst process, the variation pattern of gas emission volume during normal coal mine tunneling implicitly contains the variation pattern of gas emission volume during the outburst incubation stage. If abnormal changes that contradict this pattern can be identified, it can be determined that the coal mine tunneling face is in the formation and development stage of an outburst during that period. This invention continuously obtains real-time data on gas and wind speed environment using gas concentration sensors and wind speed sensors installed at the tunneling face, and calculates the gas emission volume. A deep analysis of the gas emission volume is conducted, and a coal and gas outburst early warning index is proposed, combining dynamic changes in gas emission volume with reverse identification based on BiLSTM gas emission volume prediction and evaluation indicators. Moving average, deviation rate, dispersion rate, volatility, RMSE, and MAPE early warning indicators are established to uncover abnormal changes in gas emission volume, thereby determining the degree of coal and gas outburst risk. The gas emission rate can be calculated in real time using existing online sensors for gas concentration and wind speed, without human interference, thus solving the problems of poor accuracy and effectiveness of traditional methods.

[0097] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the disclosure of the present invention. Attached Figure Description

[0098] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0099] Figure 1 This is a flowchart of a coal and gas outburst early warning method based on real-time on-site data provided in an embodiment of the present invention.

[0100] Figure 2 This is a schematic diagram of the coal and gas outburst early warning method based on real-time on-site data provided in this embodiment of the invention.

[0101] Figure 3 This is a schematic diagram of the principle of the bidirectional LSTM algorithm model provided in the embodiment of the present invention.

[0102] Figure 4 This is a schematic diagram of the early warning threshold obtained by calculating the comprehensive index of index differences by combining statistical indicators (moving average, deviation rate, dispersion rate, volatility) according to the embodiments of the present invention. Detailed Implementation

[0103] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0104] like Figure 1 As shown, the coal and gas outburst early warning method based on real-time on-site data provided in this embodiment of the invention includes:

[0105] S101, by continuously collecting and monitoring the gas concentration and wind speed data of the return airway of the tunneling face, and combining the cross-sectional area of ​​the return airway, the gas emission rate is calculated.

[0106] S102, the obtained gas emission data is processed by five-point cubic smoothing (using the least squares method to approximate the sampling points) to remove interference data;

[0107] S103 uses statistical methods to extract the moving average, deviation rate, dispersion rate, and volatility of gas emission change values ​​as characteristic indicators.

[0108] S104, using the methane sensor and wind speed sensor data from the T2 working face of the coal mine during two months of normal tunneling production, according to formula Q... s =60Cv s S, Q s The amount of gas emitted from the tunnel, m 3 / min; C is the gas concentration in the tunnel, %; v s S is the wind speed in the tunnel, m / s; S is the cross-sectional area of ​​the tunnel at the measurement point, m. 2The time series data of gas emission is calculated and input into a bidirectional long short-term memory recurrent neural network model for training and optimization, so that the model training error (root mean square error and mean absolute percentage error) is minimized, and a data model under normal production conditions is established.

[0109] S105, the gas emission data obtained from real-time collection and calculation is input into the data model described in step four as a test dataset to obtain the gas emission prediction value. The loss value is calculated using the difference between the prediction value and the actual value. The root mean square error (RMSE) and mean absolute percentage error (MAPE) of the loss value are used as early warning indicators for coal and gas outbursts.

[0110] S106. Based on the Analytic Hierarchy Process (AHP), construct the judgment matrix A using the 1-9 scaling method, and calculate the largest eigenvalue λ. max The consistency of the judgment matrix is ​​checked using the index weight vector W to determine the weights of each index. Based on the principle of determining the membership function, the membership function of each index is determined through a large amount of experimental data. Using the fuzzy comprehensive evaluation model, a fuzzy comprehensive evaluation set is established, and the calculated evaluation set value represents the comprehensive index of the differences of each index.

[0111] S107. Data analysis of gas emission during normal coal mine production is conducted using a comprehensive index of differences among various indicators. The maximum calculated result is multiplied by a safety factor (1.2–1.5) to determine the difference threshold. This threshold is then used to identify the calculated comprehensive index sequence of differences among various indicators, determining the degree of coal and gas outburst risk (those greater than or equal to the threshold are considered high risk).

[0112] In a preferred embodiment, step S102, the five-point cubic smoothing method, employs a time series smoothing process, utilizing a cubic least squares polynomial to smooth the time series, specifically including:

[0113] There are n+1 uniform sampling points Y1, Y2, ..., Y n-1 Y n With a sampling interval of h, perform a transformation.

[0114] μ i =(t i -t0) / h, i=0, 1, 2..., n;

[0115] Then the original nodes become μ0 = 0, μ1 = 1, μ2 = 2, ..., μ n-1 =n-1,μ n =n;

[0116] The data Y is fitted using an m-th degree polynomial. iLet the fitted polynomial be:

[0117] Y (μ) =a0+μa1+μ 2 a2+…+μ m-1 a m-1 +μ m a m

[0118] The undetermined coefficients a in the fitted polynomial i The variance sum is determined by the least squares method as follows:

[0119]

[0120] If F(a0, a1, a2, ..., a...) m The minimum value for a is 1. j (j = 0, 1, 2, ..., m) Find the partial derivative, we have

[0121]

[0122] After sorting, we get:

[0123]

[0124] Substituting n=4 and m=3 into the formula Solve for a j j = 0, 1, 2, 3, then a j Substitution Y (μ) =a0+μa1+μ 2 a2+…+μ m-1 a m-1 +μ m a m ;

[0125] Let μ n =1, 2, 3, 4, The formula for five-point cubic smoothing is... For Y i The smoothed value.

[0126] In a preferred embodiment, the data model in step S104 is a Bi-directional Long Short-Term Memory (BiLSTM) recurrent neural network model with the minimum error; the parameters of the trained Bi-directional Long Short-Term Memory (BiLSTM) recurrent neural network model are set as follows:

[0127] The batch size of data in one batch is 32.

[0128] Learning rate = 1e -4 ;

[0129] The data was executed epoch 20 times;

[0130] The data splitting ratio (split_data_ratio) is 0.833.

[0131] The number of steps is lag_num = 16;

[0132] The dimension of the input value, input_dim_size, is 16;

[0133] The predicted output value dimension output_dim_size = 1;

[0134] Hidden layer dimension hidden_dim_size = 64;

[0135] The number of LSTM layers, lstm_layer_nums, is 1.

[0136] In a preferred embodiment, in step S106, the 1-9 scale method is shown in the table.

[0137] Table 1.1-9 Scale Table

[0138]

[0139]

[0140] The technical solution of the present invention will be further described below with reference to specific embodiments.

[0141] Example

[0142] The coal and gas outburst early warning method based on real-time on-site data provided by this invention includes: analyzing the evolution process of coal and gas outbursts, proposing early warning indicators for coal and gas outbursts based on the dynamic change of gas emission and the BiLSTM gas emission prediction and evaluation index, establishing early warning indicators such as moving average, deviation rate, dispersion rate, volatility, root mean square error (RMSE), and mean absolute percentage error (MAPE), determining the weight of each indicator through the analytic hierarchy process, constructing a fuzzy comprehensive early warning model, calculating the comprehensive index of differences among each indicator, and conducting early warning for coal and gas outbursts.

[0143] The established early warning indicators were validated in a real coal and gas outburst accident. Statistical indicators, deep learning indicators, and comprehensive indicators were used to analyze the gas emission during the normal production period of the coal mine. During normal production, both statistical and deep learning indicators showed multiple orange warnings exceeding 0.5, while the comprehensive indicator did not show any warnings. This confirms that the comprehensive indicator has a significant advantage in the accuracy of gas outburst early warning and can comprehensively interpret the gas change patterns before an outburst.

[0144] Specifically, such as Figure 2 As shown, the coal and gas outburst early warning method based on real-time on-site data provided by this invention includes the following steps:

[0145] 1. Key early warning indicators based on statistics

[0146] 1) Moving average

[0147] The moving average line mainly reflects the changing trend of gas concentration time series data, indicating the dynamic engineering disturbances in the mining process and the changing state of gas concentration (rising, horizontal, or falling) within a certain period. The trend has the characteristic of stability, that is, once the trend is established, it will continue for a period of time until external forces force it to change the situation; otherwise, the trend will continue.

[0148]

[0149] In the formula, C represents the average of n gas concentration data points; n is the number of records in the sequence within a certain time period; i Let be the gas concentration of the i-th record in the sequence.

[0150] 2) Deviation rate

[0151] The deviation rate reflects the amount by which the real-time time series data of gas concentration deviates from the moving average of gas concentration for that time period. The deviation rate takes into account the changes in gas concentration during that time period due to factors such as changes in coal seam gas content, gas pressure, and coal seam permeability coefficient.

[0152]

[0153] In the formula, C t This represents the gas concentration at time t; The average value of n gas concentration data.

[0154] 3) Dispersion rate

[0155] The dispersion rate reflects the degree of dispersion of a gas concentration signal sequence, that is, the extent to which each value deviates from the mean. A larger dispersion rate indicates a greater range of gas concentration variation, and vice versa. The dispersion rate of gas concentration values ​​over a given time step indicates the degree of variation in gas emission.

[0156]

[0157] In the formula: C t The m represents the gas concentration at time t; m is the number of sequence records within a certain time period.

[0158] 4) Volatility

[0159] The description of the rate of change includes two aspects: one is the amplitude of the sequence's oscillations, which can be expressed as the rate of change of amplitude R. OSC This can be expressed as the relative rate of change between the initial and final gas emission amounts within a given time interval; another aspect is the frequency of change in the sequence, which can be represented by the frequency change rate R. OFC This refers to the number of times the methane concentration changes within a certain time interval. The combination of these two aspects describes the actual situation of methane concentration changes.

[0160] R OSC (n)=[(C Tmax -C Tmin ) / C Tmax (4)

[0161] In the formula: R osc (n) represents the rate of change of amplitude over a time interval n; C Tmax C Tmin These represent the maximum and minimum gas concentrations within the time interval.

[0162] R OFC =f(C (t-n)-t (5)

[0163] In the formula: C (t-n)-t Let f be the gas concentration from time tn to time t; f() represents the number of times the change occurs.

[0164] 2. Key Early Warning Indicators Based on Bidirectional LSTM Neural Networks (BiLSTM Model)

[0165] Bidirectional LSTM neural networks (BiLSTM) have the ability to learn using both past and future data, meaning they can use information from both the upper and lower reservoir layers to predict reservoir physical parameters, thus overcoming the limitations of traditional LSTM. A schematic diagram of the Bidirectional LSTM algorithm model is shown below. Figure 3 As shown, from bottom to top, the layers are the input layer, forward layer, backward layer, and output layer. As two LSTM network layers with opposite information transmission directions, the forward LSTM layer can acquire temporal information from previous moments, while the backward LSTM layer can acquire temporal information from future moments. This means the model can fully acquire information from both the preceding and following reservoir segments to predict the current reservoir segment's state, thus improving model performance.

[0166] The BiLSTM model uses eight weights in a loop at each time step: weights input to the forward and backward hidden layers (W1, W4); weights in the forward hidden layer (W2, W3); weights in the backward hidden layer (W6, W7); and weights from the forward and backward hidden layers to the output layer (W5, W8). The forward hidden layer h′ t Backward hidden layer h t and output layer o t The calculation formula is as follows:

[0167] h` t =f(w1x t +w2C` t-1 +w3h` t-1 (6)

[0168] h t =f(w4x t +w6C t-1 +w7h t-1 (7)

[0169] o t =g(w5h` t +w8h t (8)

[0170] Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) were selected as metrics to measure the model's prediction accuracy. RMSE and MAPE reflect the deviation between the predicted and actual gas concentrations; smaller values ​​indicate better prediction performance, while larger values ​​indicate higher distortion. MAPE is more robust to outliers and reflects the overall prediction performance of the model, while RMSE emphasizes the impact of outliers on the model's prediction performance. Using both metrics provides a better assessment of the model's prediction accuracy.

[0171] The calculation formulas are as follows:

[0172]

[0173]

[0174] In the formula: C i Indicates the actual gas concentration; This indicates the predicted gas concentration; n represents the number of samples.

[0175] 3. Construction of Coal and Gas Outburst Early Warning System

[0176] (1) Determining weights using AHP

[0177] The judgment matrix A is constructed using the 1-9 scaling method, and the largest eigenvalue λ is calculated. max And the indicator weight vector W. Consistency test of the judgment matrix. To verify the reasonableness of the allocation of evaluation indicator weights, the consistency test judgment matrix expression is:

[0178]

[0179] In the formula: C1 is the consistency ratio; n is the number of comparison indicators; R1 is the average random consistency index. When C1 < 0.1, the constructed judgment matrix meets the requirements.

[0180] (2) Determination of membership function

[0181] Based on the principles for determining membership functions, membership functions for various indicators were determined through extensive experimental data. Among these, the moving average was calculated by comparing the real-time moving average of gas concentration with the average gas concentration of the previous period (one hour). The values ​​are divided by the ratio of the real-time deviation rate and dispersion rate to the average value of the calculation period; the value range is divided by comparing the real-time volatility rate with the average value of the calculation period, and the larger value of the amplitude change rate and frequency change rate is taken; the value range is divided by the ratio of the real-time gas concentration sequence prediction result evaluation value before the accident to the average value of the normal production gas concentration sequence prediction result evaluation value.

[0182]

[0183]

[0184]

[0185]

[0186]

[0187]

[0188] (3) Establish a fuzzy comprehensive early warning model

[0189] Fuzzy comprehensive evaluation comprises three main elements: factor set, evaluation set, and single-factor evaluation. Multi-factor fuzzy comprehensive evaluation is then performed based on the single-factor evaluation.

[0190] a. Establishing a factor set

[0191] The factor set U is a set of various factors that influence the evaluation object, and can be represented as U = (u1, u2, ..., u3). n ), where element ui Let i represent the influencing factors, where i = 1, 2, ..., n. In this invention, the set of evaluation factor indicators is a set consisting of two evaluation factors: gas concentration change characteristics and gas prediction model indicators, i.e., U = (u1, u2).

[0192] b. Establish an evaluation set

[0193] The evaluation set is the set of possible evaluation results for the evaluated object, which can be represented as V = (v1, v2, ..., v...). m ), where element v j The evaluation result is represented by j = 1, 2, ..., m. In this invention, V = {an accident occurred, no accident occurred}, that is, V = {I, II}.

[0194] c. Establish weight set

[0195] The elements in the factor set U have different levels of importance in the evaluation, and each element u must be evaluated. i Different weight sets W are given according to their importance. W is a fuzzy subset of the factor set U, which can be represented as W = (w1, w2, ... w... n In this invention, since the factor set consists of two factors, the weight set is W = (w1, w2) and w1 + w2 = 1.

[0196] d. Single-factor fuzzy evaluation

[0197] First, we evaluate each factor individually to determine the degree of membership of the evaluated object to the elements of the evaluation set. Let the evaluated object be the i-th factor u in the factor set. i To make a judgment, evaluate the j-th element v in the evaluation set. j The degree of membership is r ij Then for the i-th factor u i The evaluation result can be represented as R. i ={r i1 r i2 , ...r im}, R i This represents a single-factor evaluation set.

[0198] e. Fuzzy Comprehensive Evaluation

[0199] Construct a multi-factor comprehensive evaluation (R evaluation matrix) from the single-factor evaluation set, i.e., R = {R1, R2}. T Using a weighted average model, and following the multiplication operation of fuzzy matrices, the fuzzy comprehensive evaluation set B is obtained, i.e.

[0200] B = WR = {b1, b2, ..., b} m} (18)

[0201] In the formula: b iLet be a fuzzy comprehensive evaluation index, where i = 1, 2, ..., m, representing the degree of membership of the evaluation object to the i-th element in the evaluation set V, considering all influencing factors.

[0202] The fuzzy comprehensive evaluation set B is used to represent the comprehensive index of differences among various indicators. The RMSE and MAPE indicators are predicted and evaluated using a model, and combined with the dynamic change characteristics of gas concentration, coal and gas outburst assessment and early warning are conducted. During normal coal mine production, the comprehensive index of differences among various indicators fluctuates within a certain range. When the index reaches its maximum value within the normal range, an orange warning is issued; when the comprehensive index of differences among various indicators approaches 1, it indicates a higher probability of coal and gas outburst, and a red warning is issued.

[0203] (4) Early warning of coal and gas outburst

[0204] Data analysis of gas emission during normal coal mine production periods is conducted using a comprehensive index of differences among various indicators. The maximum calculated result is multiplied by a safety factor (1.2–1.5) to determine the difference threshold. This threshold is then compared with the calculated comprehensive index sequence of differences among various indicators to determine the degree of coal and gas outburst risk. An orange alert is issued when the index reaches the threshold; a red alert is issued when the comprehensive index of differences among various indicators approaches 1, indicating a higher probability of coal and gas outburst.

[0205] The positive effects of the present invention will be further described below with reference to experimental data.

[0206] The method described above was applied to a coal mine where an accident occurred, and its effectiveness was verified. Methane and wind speed data were collected during normal production two months prior to the accident. Gas emission rates were calculated, and a trained model was used for prediction. The model's RMSE and MAPE values ​​were calculated, and a comprehensive index of indicator differences (e.g., moving average, deviation rate, dispersion rate, volatility rate) was calculated using statistical indicators. Figure 4 As shown in the figure, the warning threshold was obtained; then, the gas emission volume on the eve of the accident was used as a test set for coal and gas outburst warning. The results showed that warnings exceeding the threshold occurred 5 days before the accident, and this was verified in other coal mines where accidents occurred, consistent with the actual situation.

[0207] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the appended claims.

[0208] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure should be limited by the appended claims.

Claims

1. A coal and gas outburst early warning method based on real-time on-site data, characterized in that, The coal and gas outburst early warning method based on real-time on-site data continuously obtains real-time data on gas and wind speed by installing gas concentration sensors and wind speed sensors at the tunneling face, and calculates the gas emission rate. A deep analysis of gas emission volume was conducted to obtain early warning indicators for coal and gas outbursts that combine dynamic changes in gas concentration with reverse identification based on gas emission volume prediction and evaluation indicators. Early warning indicators such as moving average, deviation rate, dispersion rate, volatility, root mean square error, and mean absolute percentage error were established. The weights of each indicator are determined by the analytic hierarchy process, a fuzzy comprehensive early warning model is constructed, and the comprehensive index of the differences among the indicators is calculated to provide early warning for coal and gas outbursts. The coal and gas outburst early warning method based on real-time on-site data specifically includes: Step 1: By continuously collecting and monitoring the gas concentration and wind speed data of the return airway at the tunneling face, and combining the cross-sectional area of ​​the return airway, the gas emission rate is calculated. Step 2: Perform a five-point triple smoothing process on the obtained gas emission data to remove interference data; Step 3: Use statistical methods to extract the moving average, deviation rate, dispersion rate, and volatility of the gas emission change value as characteristic indicators; Step four: Collect methane and wind speed data from the working face of the coal mine during two months of normal tunneling production, and then, according to the formula... The time series data of gas emission were calculated, where Q s The amount of gas emitted from the tunnel, m 3 / min; C is the methane concentration in the tunnel, %; v s S is the wind speed in the tunnel, m / s; S is the cross-sectional area of ​​the tunnel at the measurement point, m. 2 The calculated gas emission time-series data is input into a bidirectional long short-term memory recurrent neural network model for training and optimization to minimize the model training error and establish a data model under normal production conditions. This data model is the bidirectional long short-term memory recurrent neural network model with the smallest error. The parameters of the bidirectional long short-term memory recurrent neural network model with the smallest error established after training are set as follows: The batch size is 32. learning rate = 1e -4 ; The data was executed epoch-timed 20 times. The data splitting ratio is split_data_ratio = 0.833; The number of steps is lag_num=16; The dimension of the input value is input_dim_size=16; The predicted output value dimension output_dim_size=1; Hidden layer dimension hidden_dim_size=64; The number of LSTM layers is set to lstm_layer_nums=1; Step 5: Use the real-time collected and calculated gas emission data as a test dataset and input it into the data model in Step 4 to obtain the predicted gas emission value. Calculate the loss value using the difference between the predicted value and the actual value, and use the root mean square error and mean absolute percentage error of the loss value as early warning indicators for coal and gas outbursts. Step six: Using the analytic hierarchy process (AHP), compare each indicator pairwise according to its frequency of appearance in the literature. ij To compare the importance of element i and element j, a matrix is ​​constructed using the 1-9 scaling method, called the judgment matrix; the largest eigenvalue λ is then calculated. max The consistency of the judgment matrix is ​​checked using the index weight vector W to determine the weights of each index. Based on the principle of determining the membership function, the membership function of each index is determined through experimental data. Using the fuzzy comprehensive evaluation model, a fuzzy comprehensive evaluation set is established, and the calculated evaluation set values ​​represent the comprehensive index of the differences of each index. Step 7: Analyze the gas emission during normal coal mine production using the comprehensive index of differences among various indicators. Multiply the maximum value of the calculated result by the safety factor to determine the difference threshold. Use this threshold to identify the calculated comprehensive index sequence of differences among various indicators and determine the degree of coal and gas outburst risk.

2. The coal and gas outburst early warning method based on real-time on-site data as described in claim 1, characterized in that, In step two, the five-point cubic smoothing method employs a time series smoothing technique, utilizing a cubic least squares polynomial to smooth the time series. Specifically, this includes: There are n+1 uniform sampling points Y1, Y2, ..., Y n-1 ,Y n With a sampling interval of h, perform a transformation. ,i=0,1,2……,n; Then the original nodes become μ0=0, μ1=1, μ2=2, ..., μ n-1 =n-1,μ n =n; The data Y is fitted using an m-th degree polynomial. i Let the fitted polynomial be: ， The undetermined coefficients a in the fitted polynomial i The variance sum is determined by the least squares method as follows: , If F(a0,a1,a2,……,a) m The minimum value for a is 1. j (j=0,1,2,……,m) Find the partial derivatives, we have , After sorting, we get: , Substituting n=4 and m=3 into the formula Solve for a j , 𝑗=0,1,2,3, then a j Substitution ; The formula for five-point cubic smoothing is... For Y i The smoothed value.

3. The coal and gas outburst early warning method based on real-time on-site data as described in claim 1, characterized in that, In step three, statistical methods are used to extract the moving average, deviation rate, dispersion rate, and volatility of the gas emission rate as characteristic indicators, specifically including: 1) Extract the moving average of the gas emission rate to obtain the trend of gas concentration time series data. The formula for calculating the average value of gas concentration data is: In the formula, C represents the average of n gas concentration data points; n is the number of records in the sequence within a certain time period; i Let be the gas concentration of the i-th record in the sequence; 2) Extract the deviation rate of gas emission change values ​​to obtain the amount by which the real-time time series data of gas concentration deviates from the moving average of gas concentration for that time period; the deviation rate is calculated using the following formula: In the formula, C t This represents the gas concentration at time t; The average value of n gas concentration data; 3) Extract the dispersion rate of the gas emission rate change value to obtain the degree of dispersion of the gas concentration signal sequence. The dispersion rate calculation formula is: In the formula: C t The m represents the gas concentration at time t; m is the number of sequence records within a certain time period. 4) Extract the volatility of gas emission changes to obtain the actual situation of gas concentration changes; volatility extraction includes calculation of amplitude change rate and frequency change rate. Formula for calculating the rate of change of amplitude R OSC for: In the formula: Let n be the rate of change of amplitude over a time interval. , These represent the maximum and minimum gas concentrations within the time interval; Formula for calculating frequency change rate R OFC for: In the formula: The methane concentration is measured from time tn to time t. This indicates the number of changes calculated.

4. The coal and gas outburst early warning method based on real-time on-site data as described in claim 1, characterized in that, In step four, the bidirectional long short-term memory recurrent neural network model includes: At each time step, eight weights are used in a cycle: weights input to the forward and backward hidden layers (W1, W4); weights of the forward hidden layer (W2, W3); weights of the backward hidden layer (W6, W7); weights from the forward and backward hidden layers to the output layer (W5, W8); weights of the forward hidden layer... Backward hidden layer h t and output layer o t The calculation formulas are as follows: 。 5. The coal and gas outburst early warning method based on real-time on-site data as described in claim 1, characterized in that, The formula for calculating the root mean square error in step five is as follows: , The formula for calculating the mean absolute percentage error is: In the formula: C i Indicates the actual gas concentration; This indicates the predicted gas concentration; n represents the number of samples.

6. The coal and gas outburst early warning method based on real-time on-site data as described in claim 1, characterized in that, In step six, determining the weights of each indicator specifically includes: constructing a judgment matrix A using the 1-9 scaling method, and calculating the largest eigenvalue λ. max And the indicator weight vector W, the consistency test of the judgment matrix; the expression of the consistency test judgment matrix is: In the formula: C1 is the consistency ratio; n is the number of comparison indicators; R1 is the average random consistency index; when C1 < 0.1, the constructed judgment matrix meets the requirements.

7. The coal and gas outburst early warning method based on real-time on-site data as described in claim 1, characterized in that, In step six, the membership function determination includes: Moving average: Utilizes the real-time moving average of gas concentration and the average gas concentration of the previous period. Divide by the ratio; Deviation rate and dispersion rate: The value range is divided by comparing the real-time deviation rate and dispersion rate with the average value of the calculation period; Volatility: The range of values ​​is divided by comparing real-time volatility with the average value of the calculation period; Root mean square error and mean absolute percentage error: The range of values ​​is divided by the ratio of the evaluation value of the gas concentration sequence prediction result before the real-time accident to the evaluation average value of the gas concentration sequence prediction result during normal production. 。 8. The coal and gas outburst early warning method based on real-time on-site data as described in claim 1, characterized in that, In step six, the establishment of the fuzzy comprehensive evaluation model includes: a. Establishing a factor set The factor set U is represented as ; where element u i Indicate the influencing factors, among which ; b. Establish an evaluation set The evaluation set V is denoted as , where element v j This indicates the evaluation result, in which When V = {an accident occurs, no accident occurs}, ; c. Establish weight set The elements in the factor set U have different levels of importance in the evaluation, and each element u... i Different weight sets W are given according to their importance, where W is a fuzzy subset of the factor set U, denoted as: The weight set is hour, ; d. Single-factor fuzzy evaluation The evaluation object is the i-th factor u in the factor set. i To make a judgment, evaluate the j-th element v in the evaluation set. j The degree of membership is r ij Then for the i-th factor u i The evaluation result is expressed as , Represents a single-factor evaluation set; e. Fuzzy Comprehensive Evaluation Construct a multi-factor comprehensive evaluation matrix R from the single-factor evaluation set. Using a weighted average model and following the multiplication operation of fuzzy matrices, the fuzzy comprehensive evaluation set B is obtained as follows: ； In the formula: b i The evaluation index is a fuzzy comprehensive evaluation index, among which, , which represents the degree of membership of the evaluated object to the i-th element in the evaluation set V after comprehensively considering all influencing factors.

9. The coal and gas outburst early warning method based on real-time on-site data as described in claim 1, characterized in that, In step seven, the value of the fuzzy comprehensive evaluation set B is used to represent the comprehensive index of the differences between the indicators. The root mean square error and the mean absolute percentage error of the evaluation indicators are predicted by the model. Combined with the dynamic change characteristics of gas concentration, coal and gas outburst evaluation and early warning are carried out. When the coal mine is in normal production, the comprehensive index of the differences between the indicators fluctuates within the normal range. When the index reaches the maximum value within the normal range, an orange warning is issued. When the composite index of the differences among various indicators approaches 1, it indicates a greater likelihood of coal and gas outbursts, and a red alert is issued.

Images