An Adaptive Long-Term Prediction Method and System for Gas Overlimit

Abstract

This application relates to the technical field of coal mine gas overrun prediction, and provides an adaptive long-term gas overrun prediction method and system. In this method, the multi-source data obtained from the coal mine underground is processed to construct the characteristics of the data at different time scales; according to the Pearson correlation coefficients of the characteristics at different time scales with the target gas concentration, the best prediction index combination is adaptively selected; the optimized parameter combination of the Informer long-term prediction model is determined, and the Informer long-term prediction model is optimized based on the grey wolf optimization algorithm to obtain the best prediction model for long-term overrun prediction of the gas concentration in the working face, and the prediction results are comprehensively evaluated based on the pre-determined accuracy evaluation index. Thereby, it effectively solves the problems in the prior art that the gas overrun disaster-causing factors vary greatly in different mining areas, different coal mines, and different working faces, with weak generalization ability and insufficient long-term prediction ability.

Description

Technical Field

[0001] The present application relates to the technical field of coal mine gas overrun prediction, and particularly relates to an adaptive long-term gas overrun prediction method and system. Background Technique

[0002] Coal mine gas disasters are one of the major disasters threatening coal mine safety production. Gas disasters directly hinder the normal production of coal mines and impede the sustainable, stable and healthy development of the coal industry. Strengthening gas disaster prevention and control is an important guarantee to ensure the stable and reliable supply of coal energy and promote the all-round and healthy development of the national economy.

[0003] With the high attention to coal mine safety production and the need for the self-development of coal mine enterprises, the accurate prediction of gas concentration, the disaster forecast and prevention of gas outburst are of great significance. Traditional gas prediction methods are determined based on quantitative indicators of the properties of gas-bearing coal seams and their occurrence conditions, such as coal seam property indicators, gas indicators, ground stress indicators or comprehensive indicators, to predict whether one or more of these indicators exceed the critical value. However, the prediction accuracy of traditional prediction methods is difficult to meet the requirements of coal mine safety production.

[0004] Currently, for the prediction of coal mine gas concentration, one method is to construct a parameter equation by simulating the diffusion law of gas in the coal mining process and solve it. The prediction result of this method is too idealized and cannot actually reflect the complex environment in the coal mine underground. Another method is to construct a prediction model through feature engineering or deep learning technology to predict the gas concentration. However, the features extracted by this type of method cannot fully express the change trend of gas concentration in different environments, resulting in low prediction accuracy.

[0005] Therefore, there is an urgent need to provide a technical solution to address the deficiencies of the above-mentioned existing technologies. Summary of the Invention

[0006] The purpose of the present application is to provide an adaptive long-term gas overrun prediction method and system to solve or alleviate the problems existing in the above-mentioned existing technologies.

[0007] To achieve the above purpose, the present application provides the following technical solutions:

[0008] The present application provides an adaptive long - term prediction method for gas exceeding the limit, including: Step S101, processing the obtained multi - source data {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} in the coal mine underground to construct different time - scale features of the multi - source data in the coal mine underground; where T0 is the gas concentration at the upper corner of the working face, T1 is the gas concentration of the working face, T2 is the gas concentration of the return airway, T7 is the gas concentration of the refuge chamber, FS is the air velocity of the working face, W is the temperature of the working face, GT is the gas drainage concentration in the goaf, GL is the gas drainage flow in the goaf, GW is the gas drainage temperature in the goaf, and GP is the gas drainage pressure in the goaf;

[0009] Step S102, adaptively screening the best prediction index combination of the gas concentration T1 of the working face according to the Pearson correlation coefficient between the gas concentration T1 of the working face and the different time - scale features; Step S103, optimizing the optimization parameter combination of the pre - constructed Informer long - term prediction model based on the Grey Wolf Optimization algorithm to obtain the best prediction model; Step S104, based on the best prediction model, performing over - limit prediction on the gas concentration T1 of the working face, and comprehensively evaluating the prediction results based on the pre - determined accuracy evaluation index.

[0010] Preferably, in Step S101, based on the Holt exponential smoothing algorithm, missing value processing is performed on the obtained multi - source data in the coal mine underground; the multi - source data in the coal mine underground after missing value processing is standardized, and outlier identification is performed on the standardized data based on the 3σ method; based on the autoregressive model, outliers identified in the multi - source data in the coal mine underground are processed; for the multi - source data in the coal mine underground after outlier processing, the maximum value Y t，max 、minimum value Y t，min 、average value Y t，ave 、difference value Y t，dif are constructed respectively for 5 min, 30 min, 1 h, and 24 h.

[0011] Preferably, according to the autoregressive model:

[0012]

[0013] outliers identified in the multi - source data in the coal mine underground are processed; where M is the outlier at the i - th moment in the multi - source data in the coal mine underground; c is the constant term; p is the autoregressive order, representing the time interval; is the autocorrelation coefficient; ε is the random error value; X i-j is the historical observation value at the j - th moment before the i - th moment in the multi - source data.

[0014] Preferably, according to the formula:

[0015]

[0016] For the multi-source data in the coal mine underground after outlier processing, the maximum values Y of 5 min, 30 min, 1 h, and 24 h are extracted respectively t，max , the minimum value Y t，min , the average value Y t，ave , the difference value Y t，dif ; where t represents different time scales; x t represents the multi-source data in the coal mine underground after outlier processing within the t time range; x t-1 represents the multi-source data in the coal mine underground after outlier processing within the previous t time range.

[0017] Preferably, in step S102, it includes: calculating the Pearson correlation coefficient between the gas concentration T1 of the working face and the characteristics of different time scales, and adaptively extracting the gas overrun prediction indicators whose absolute value of the Pearson correlation coefficient is in [0.4, 1] to form the optimal prediction indicator combination.

[0018] Preferably, in step S103, determining the number of layers of the multi-head probability sparse self-attention mechanism, the number of layers of self-attention distillation, and the number of fully connected layers of the Informer long-term prediction model as the optimized parameter combination of the Informer long-term prediction model.

[0019] Preferably, in step S103, the Informer long-term prediction model includes: an encoder and a decoder; the encoder is composed of 3 identical layers stacked, and each layer contains a multi-head probability sparse self-attention mechanism sub-layer and a self-attention distillation sub-layer; the decoder is composed of 2 identical layers stacked, and each layer contains a masked multi-head probability sparse self-attention mechanism sub-layer and a self-attention mechanism sub-layer.

[0020] Preferably, in step S104, according to the formula:

[0021]

[0022] Calculate the mean absolute error MAE, mean square error MSE, root mean square error RMSE, and determination coefficient R between the prediction result obtained by the optimal prediction model and the true value of the gas concentration of the target working face 2 ; where H k is the k-th true value of the gas concentration of the working face; h k is the prediction result corresponding to the k-th true value of the gas concentration; J k is the mean value of the K true values of the gas concentration of the working face; k ∈ K, both k and K are positive integers.

[0023] The embodiment of the present application also provides an adaptive long-term prediction system for gas overrun, including:

[0024] A source data unit configured to process the acquired multi-source data {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} in the coal mine underground to construct different time-scale features of the multi-source data in the coal mine underground; where T0 is the gas concentration at the upper corner of the working face, T1 is the gas concentration of the working face, T2 is the gas concentration of the return airway, T7 is the gas concentration of the refuge chamber, FS is the wind speed of the working face, W is the temperature of the working face, GT is the gas drainage concentration in the goaf, GL is the gas drainage flow in the goaf, GW is the gas drainage temperature in the goaf, and GP is the gas drainage pressure in the goaf;

[0025] An index unit configured to adaptively screen the best prediction index combination of the gas concentration of the working face according to the Pearson correlation coefficient of the different time-scale features of the gas concentration of the working face;

[0026] A parameter optimization unit configured to optimize the optimized parameter combination of the pre-constructed Informer long-term prediction model based on the grey wolf optimization algorithm to obtain the best prediction model;

[0027] A comprehensive evaluation unit configured to perform overrun prediction on the gas concentration of the working face based on the best prediction model, and perform a comprehensive evaluation on the prediction result based on the pre-determined accuracy evaluation index.

[0028] Beneficial effects: An adaptive long-term prediction method for gas overrun provided by the present application processes the acquired multi-source sensor data in the coal mine underground, constructs different time-scale features of the multi-source data in the coal mine underground, and adaptively screens the best prediction index combination of the gas concentration of the working face according to the Pearson correlation coefficient between the gas concentration T1 of the coal mine working face and the different time-scale features; then, based on the grey wolf optimization algorithm, the optimized combination parameters of the pre-constructed Informer long-term prediction model are optimized to obtain the best prediction model; finally, based on the best prediction model, overrun prediction is performed on the gas concentration T1 of the working face, and a comprehensive evaluation is performed on the prediction result based on the pre-determined accuracy evaluation index. Thereby, it effectively solves the problems in the prior art that the disaster-causing factors of gas overrun in different mining areas, different coal mines, and different working faces vary greatly, the generalization ability is weak, and the long-term prediction ability is insufficient. It adaptively selects the best prediction index for different mining areas, different coal mines, and different working faces, and constructs the long-term scale features of gas overrun, so as to predict the gas concentration of the working face, greatly improve the long-term prediction effect of the gas concentration overrun of the working face, increase the overrun disposal time, and is particularly suitable for gas overrun prediction in high-gas mines. Description of the Drawings

[0029] The accompanying drawings of the specification, which form a part of this application, are used to provide a further understanding of this application. The schematic embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation to this application. Among them:

[0030] Figure 1 It is a schematic flow chart of an adaptive long-term prediction method for gas overrun according to some embodiments of this application;

[0031] Figure 2 It is a schematic framework diagram of an Informer long-term prediction model according to some embodiments of this application;

[0032] Figure 3 It is a schematic diagram in the form of a heat map of the Pearson correlation coefficient r between the gas concentration T1 of the working face and different time-scale features according to some embodiments of this application;

[0033] Figure 4 It is a schematic structural diagram of an adaptive long-term prediction system for gas overrun according to some embodiments of this application. Detailed implementation manners

[0034] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. Each example is provided by way of explanation of the present application rather than limitation thereof. In fact, those skilled in the art will clearly understand that modifications and variations can be made to the present application without departing from the scope or spirit of the present application. For example, features shown or described as part of one embodiment can be used in another embodiment to yield yet another embodiment. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the embodiments of the present invention shall fall within the scope of protection of the embodiments of the present invention.

[0035] Gas accidents are the main factors threatening the safe production in coal mines. Real-time and accurate monitoring of gas concentration is an important means to prevent gas accidents. In the safe production of coal mines, accurate and timely gas concentration prediction can effectively reduce and prevent coal mine gas accidents.

[0036] Gas overrun is one of the main inducements for gas accidents. The coal mining working face in a coal mine is a place where gas overrun occurs frequently. However, the disaster-causing factors of gas overrun are complex and are not only closely related to objective factors such as gas occurrence characteristics, gas geology, coal and rock stress, and coal body structure, but also affected by human factors such as mine excavation deployment and layout, ventilation mode, coal mining technology, operation mode, and personnel quality. At the same time, the factors causing gas overrun in each mining area, each coal mine, and even each working face are different.

[0037] At present, most coal mines only have the function of gas overrun alarm, but lack the function of gas overrun prediction. For the small number of coal mines with the function of gas overrun prediction, there are also problems such as single prediction index, weak generalization ability, and poor prediction effect, and they cannot automatically adapt to the characteristics of each mining area, each coal mine, and each working face. In addition, the existing methods can only perform short-term (5-15 minutes) prediction, and have large prediction errors and poor prediction effects for long-term (more than 1 hour) prediction. Therefore, on the basis of comprehensively considering the characteristics and differences of each mining area, each coal mine, and each working face, adaptively select prediction indicators, deeply mine the time series characteristics of gas concentration, and conduct long-term prediction of gas overrun, which can move the technical checkpoint of gas overrun prevention forward, increase the emergency decision-making time, and is of great significance for preventing coal mine gas and ensuring coal mine safety production.

[0038] Based on this, the embodiments of the present application propose an adaptive long-term gas overrun prediction method and system, which adaptively select the optimal prediction indicators according to the characteristics of each mining area, each coal mine, and each working face, and extract different time-scale characteristics of gas overrun, so as to predict the gas concentration in advance, greatly improve the long-term prediction effect of gas overrun, increase the overrun disposal time, and are applicable to mine gas overrun prediction, especially to gas overrun prediction in high-gas mines.

[0039] As Figures 1 to 3 shown, the adaptive long-term gas overrun prediction method includes:

[0040] Step S101: Process the obtained multi-source data {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} in the coal mine underground to construct different time-scale characteristics of the multi-source data in the coal mine underground.

[0041] In the present application, the multi-source data is obtained through multi-source sensors in the coal mine underground. Specifically, it includes: the gas concentration T0 at the upper corner of the working face, the gas concentration T1 of the working face, the gas concentration T2 in the return airway, the gas concentration T7 in the refuge chamber, the working face air velocity FS, the working face temperature W, the gas extraction concentration GT in the goaf, the gas extraction flow GL in the goaf, the gas extraction temperature GW in the goaf, and the gas extraction pressure GP in the goaf. Here, it should be noted that the obtained multi-source data is a data set containing {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} within a certain time period, that is, the multi-source data contains multiple groups of {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} at the same moment.

[0042] After obtaining the multi-source data in the coal mine underground, preprocess the multi-source data in the coal mine underground, specifically including: missing value processing, standardization processing, and outlier processing. Among them, based on the Holt exponential smoothing algorithm, perform missing value processing on the obtained multi-source data in the coal mine underground, specifically according to the formula:

[0043]

[0044] α and β are the built-in smoothing parameters of the Holt exponential smoothing method, and can take 0.5. X i is the missing value at time i, and b i and b i-1 are the trend values at time i and time i - 1 respectively. S i and S i-1 are the smoothed values at time i and time i - 1 respectively. L is the prediction step, which takes 5 here.

[0045] Then, the multi-source data in the coal mine underground after missing value processing is standardized. Specifically, according to the maximum-minimum standardization:

[0046]

[0047] The multi-source data in the coal mine underground after missing value processing is standardized. Among them, y is the value after standardizing the multi-source data in the coal mine underground; x' is the value after filling the multi-source data in the coal mine underground with the Holt exponential smoothing algorithm; max(x′) is the maximum value after filling the multi-source data in the coal mine underground with the Holt exponential smoothing algorithm; min(x′) is the minimum value after filling the multi-source data in the coal mine underground with the Holt exponential smoothing algorithm.

[0048] By standardizing the multi-source data, the influence of the data dimension difference between different indicators is effectively eliminated. Furthermore, based on the 3σ method, the outliers in the standardized data are identified. Specifically, if the multi-source data follows a normal distribution with a mean of u and a standard deviation of σ, the data outside the range of (u - 3σ, u + 3σ) can be considered as outliers.

[0049] Next, through the autoregressive model:

[0050]

[0051] The outliers identified in the multi-source data in the coal mine underground are processed. Among them, M is the outlier at time i in the multi-source data in the coal mine underground; c is the constant term; p is the autoregressive order, indicating the time interval; is the autocorrelation coefficient; ε is the random error value, indicating a white noise signal with a mean of 0 and a constant standard deviation of s; X i-j is the historical observation value at time j before time i in the multi-source data.

[0052] Finally, according to the formula:

[0053]

[0054] For the multi-source data in the coal mine after outlier processing, extract the maximum value Y, minimum value Y, average value Y, and difference value Y at 5 minutes, 30 minutes, 1 hour, and 24 hours respectively. Here, t represents different time scales; x represents the multi-source data in the coal mine after outlier processing within the t time range; x represents the multi-source data in the coal mine after outlier processing within the previous t time range. t，max , minimum value Y t，min , average value Y t，ave , difference value Y t，dif . In the formula, t represents different time scales; x t represents the multi-source data in the coal mine after outlier processing within the t time range; x t-1 represents the multi-source data in the coal mine after outlier processing within the previous t time range.

[0055] Here, it should be noted that the multi-source data includes multiple groups of {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} at the same moment. When preprocessing the data, each item in multiple groups of {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} at the same moment is preprocessed separately. Similarly, the constructed features of different time scales include the constructed features of the gas concentration T0 at the upper corner of the working face {Y t，max , Y t，min , Y t，ave , Y t，dif}, the constructed features of the gas concentration T1 at the working face {Y t，max , Y t，min , Y t，ave , Y t，dif}, the constructed features of the gas concentration T2 in the return airway {Y t，max , Y t，min , Y t，ave , Y t，dif}, the constructed features of the gas concentration T7 in the refuge chamber {Y t，max , Y t，min , Y t，ave , Y t，dif}, the constructed features of the wind speed FS {Y t，max , Y t，min , Y t，ave , Y t，dif}, the constructed features of the temperature W {Y t，max , Y t，min , Y t，ave , Y t，dif}, the constructed features of the gas drainage concentration GT in the goaf {Y t，max , Y t，min , Y t，ave , Y t，dif}, the constructed features of the gas drainage flow GL in the goaf {Y t，max , Y t，min , Y t，ave , Y t，dif}, and the constructed features of the gas drainage gas temperature GW in the goaf {Y t，max, Y t，min , Y t，ave , Y t，dif}, Structural characteristics of the goaf gas drainage pressure GP{Y t，max , Y t，min , Y t，ave , Y t，dif}.

[0056] Step S102: According to the Pearson correlation coefficient between the working face gas concentration T1 and the characteristics of different time scales, adaptively screen the optimal prediction index combination of the working face gas concentration T1.

[0057] The Pearson correlation coefficient is a statistic that describes the strength of the linear correlation between two variables. According to the formula:

[0058]

[0059] Calculate the Pearson correlation coefficient r between the working face gas concentration T1 and the characteristics of different time scales. In the formula, X j is the main index, that is, the working face gas concentration T1, and Y j is the constructed index after the multi-source data is constructed with different time scales, is the average value of the main index X j , is the average value of the characteristic index Y j , and Q is the total number of characteristic variables.

[0060] In this application, the structural characteristics of the gas concentration T0 at the upper corner of the working face{Y t，max , Y t，min , Y t，ave , Y t，dif}, the structural characteristics of the working face gas concentration T1{Y t，max , Y t，min , Y t，ave , Y t，dif}, the structural characteristics of the return airway gas concentration T2{Y t，max , Y t，min , Y t，ave , Y t，dif}, the structural characteristics of the gas concentration T7 in the refuge chamber{Y t，max , Y t，min , Y t，ave , Y t，dif}, the structural characteristics of the wind speed FS{Y t，max , Y t，min , Y t，ave , Y t，dif}, the structural characteristics of the temperature W{Y t，max , Y t，min , Y t，ave , Y t，dif}, Structural characteristics of the gas drainage concentration GT in the gob area {Y t，max , Y t，min , Y t，ave , Y t，dif}, Structural characteristics of the gas drainage flow rate GL in the gob area {Y t，max , Y t，min , Y t，ave , Y t，dif}, Structural characteristics of the gas drainage temperature GW in the gob area {Y t，max , Y t，min , Y t，ave , Y t，dif}, Structural characteristics of the gas drainage pressure GP in the gob area {Y t，max , Y t，min , Y t，ave , Y t，dif} As structural indicators, calculate the Pearson correlation coefficient r between the gas concentration T1 at the working face and the characteristics of different time scales, that is, calculate the Pearson correlation coefficient between the gas concentration T1 at the working face and different structural indicators.

[0061] That is, calculate the Pearson correlation coefficient between the gas concentration T1 at the working face and each item in the structural indicators. For example, calculate the Pearson correlation coefficient between the gas concentration T1 at the working face and the structural characteristics of the gas concentration T0 at the upper corner of the working face {Y t，max , Y t，min , Y t，ave , Y t，dif}, between the gas concentration T1 at the working face and the structural characteristics of the gas concentration T1 at the working face {Y t，max , Y t，min , Y t，ave , Y t，dif}, between the gas concentration T1 at the working face and the structural characteristics of the gas concentration T2 in the return airway {Y t，max , Y t，min , Y t，ave , Y t，dif}, and between the gas concentration T1 at the working face and the structural characteristics of the gas drainage concentration GT in the gob area {Y t，max , Y t，min , Y t，ave , Y t，dif}.

[0062] The Pearson correlation coefficient r ranges from [-1, 1]. Two variables with an absolute value of r between [0.1, 0.3] are called weakly correlated, those with an absolute value between [0.4, 0.7] are called moderately correlated, and those with an absolute value between [0.8, 1] are called strongly correlated. In this application, according to the Pearson correlation coefficient r, medium and strong correlation indicators are adaptively extracted as the best prediction indicator combination for the long-term prediction model, that is, the gas overrun prediction indicators with an absolute value of the Pearson correlation coefficient in the range of [0.4, 1] are extracted to form the best prediction indicator combination.

[0063] Here, it should be noted that the gas overrun prediction indicators are the structural indicators, that is, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the gas concentration T0 at the upper corner of the working face, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the gas concentration T1 on the working face, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the gas concentration T2 in the return airway, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the gas concentration T7 in the refuge chamber, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the wind speed FS, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the temperature W, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the gas drainage concentration GT in the goaf, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the gas drainage flow GL in the goaf, the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif} of the gas drainage temperature GW in the goaf, and the structural characteristics {Y t，max 、Y t，min 、Y t，ave 、Y t，dif}. Extract the gas overlimit prediction indicators with the absolute value of the Pearson correlation coefficient in the range of [0.4, 1] from the Pearson correlation coefficients between the calculated gas concentration t1 at the working face and each item in the structural indicators to form the best prediction indicator combination.

[0064] Step S103: Based on the Grey Wolf Optimization algorithm, optimize the optimization parameter combination of the pre-constructed Informer long-term prediction model to obtain the best prediction model.

[0065] In this application, the Informer long-term prediction model is built using the PyTorch open-source deep learning framework, mainly including: an encoder (Encoder) and a decoder (Decoder). Among them, the encoder consists of 3 identical layers stacked, and each layer contains a multi-head probabilistic sparse self-attention mechanism sub-layer and a self-attention distillation sub-layer; the decoder consists of 2 identical layers stacked, and each layer contains a masked multi-head probabilistic sparse self-attention mechanism sub-layer and a self-attention mechanism sub-layer. The encoder extracts the long-term dependence relationship of the time features of multi-source sensor data (multi-source data), and the decoder is mainly used to achieve long-term prediction.

[0066] Determine the best parameter combination by continuously modifying the parameters of the Informer long-term prediction model, and then obtain the best prediction model. Specifically, determine the number of layers of the multi-head probabilistic sparse self-attention mechanism, the number of layers of self-attention distillation, and the number of fully connected layers of the Informer long-term prediction model as the optimization parameter combination of the Informer long-term prediction model.

[0067] In a specific example, first, initialize the positions and population size of the grey wolf population in the Grey Wolf Optimization algorithm. The number of grey wolf populations is set to 50, and the mean square error MSE of the Informer model is used as the fitness function. Then, iteratively calculate the fitness of 50 grey wolf individuals for 50 times, update the positions of the grey wolf individuals according to the fitness, compare the current fitness of the model with the historical optimal fitness, output the best grey wolf position and the best fitness, and then obtain the optimized best parameter combination, that is, the optimization parameter combination.

[0068] Step S104: Based on the best prediction model, perform overlimit prediction on the gas concentration T1 at the working face, and comprehensively evaluate the prediction results based on the pre-determined accuracy evaluation indicators.

[0069] After obtaining the best prediction model of Informer, the best prediction model is deployed to the edge box (i.e., the edge server) at the working face in the coal mine. The edge box is connected to the monitoring sub-station and the monitoring sensors in sequence. Through the established multi-source data real-time acquisition module, the multi-source data {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} is transmitted in real-time or at regular intervals, and the incoming multi-source data is sequentially processed for missing value handling, standardization, outlier handling, different time-scale feature construction, adaptive index screening, etc. The model receives the processed data and makes predictions, outputting the gas concentration T1 at the working face in the next 1h. The gas concentration T1 at the working face in the next 1h is compared with the safety threshold of the gas concentration T1 set for the coal mine. If it exceeds the safety threshold, an audible and visual alarm signal is issued.

[0070] In this application, by performing over-limit prediction on the gas concentration T1 at the working face and based on the gas concentration monitoring data in the coal mine, the accuracy of the prediction results of the Informer long-term prediction model is evaluated. Specifically, according to the formula:

[0071]

[0072] Calculate the mean absolute error MAE, mean square error MSE, root mean square error RMSE, and coefficient of determination R between the prediction results obtained by the best prediction model and the true values of the gas concentration at the target working face. 2 ; where H k is the k-th true value of the gas concentration at the working face, that is, the gas concentration at the working face obtained by monitoring; h k is the prediction result corresponding to the k-th true value of the gas concentration, that is, the gas concentration at the working face predicted by the Informer long-term prediction model; J k is the mean value of the K true values of the gas concentration at the working face; k ∈ K, and both k and K are positive integers.

[0073] In this application, the multi-source data obtained in the coal mine is processed through the Holt exponential smoothing algorithm, maximum-minimum standardization, autoregressive method, etc., and data is specifically extracted to construct features; the best prediction indicators are adaptively selected according to the Pearson correlation coefficient; the Informer time series prediction model constructed based on the encoder (Encoder), decoder (Decoder), and fully connected layer can further capture the long-term dependence relationship of the data, realize long-term prediction of gas over-limit, is flexible and reliable, has strong prediction ability, and is particularly suitable for gas over-limit prediction in high-gas mines, especially for gas over-limit prediction in intelligent coal mining working faces.

[0074] The embodiment of this application also provides an adaptive long-term gas over-limit prediction system, as Figure 4 shown. This prediction system includes:

[0075] The source data unit 401 is configured to process the acquired multi-source data {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} in the coal mine underground to construct different time-scale features of the multi-source data in the coal mine underground; where T0 is the gas concentration at the upper corner of the working face, T1 is the gas concentration of the working face, T2 is the gas concentration of the return airway, T7 is the gas concentration of the refuge chamber, FS is the full working face wind speed, W is the working face temperature, GT is the gas drainage concentration in the goaf, GL is the gas drainage flow rate in the goaf, GW is the gas drainage temperature in the goaf, and GP is the gas drainage pressure in the goaf;

[0076] The index unit 402 is configured to adaptively screen the optimal prediction index combination of the gas concentration of the working face according to the Pearson correlation coefficient of the different time-scale features of the gas concentration in the working face;

[0077] The parameter optimization unit 403 is configured to optimize the optimized parameter combination of the pre-constructed Informer long-term prediction model based on the grey wolf optimization algorithm to obtain the best prediction model;

[0078] The comprehensive evaluation unit 404 is configured to perform over-limit prediction on the gas concentration of the working face based on the best prediction model, and comprehensively evaluate the prediction result based on the pre-determined accuracy evaluation index.

[0079] The adaptive gas over-limit long-term prediction system provided by the embodiments of the present application can implement the steps and processes of the adaptive gas over-limit long-term prediction method in any of the above embodiments, and achieve the same technical effects, which will not be elaborated herein one by one.

[0080] In the description of the present invention, the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0081] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An adaptive long-term prediction method for gas overrun, characterized in that, Including: Step S101: Process the obtained multi-source data {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} in the coal mine underground to construct different time-scale features of the multi-source data in the coal mine underground; where T0 is the gas concentration at the upper corner of the working face, T1 is the gas concentration of the working face, T2 is the gas concentration of the return airway, T7 is the gas concentration of the refuge chamber, FS is the air velocity of the working face, W is the temperature of the working face, GT is the gas drainage concentration in the goaf, GL is the gas drainage flow in the goaf, GW is the gas drainage temperature in the goaf, and GP is the gas drainage pressure in the goaf; Step S102: According to the Pearson correlation coefficient between the gas concentration T1 of the working face and the different time-scale features, adaptively screen the best prediction index combination of the gas concentration T1 of the working face; Step S103: Based on the Grey Wolf Optimization Algorithm, optimize the optimization parameter combination of the pre-constructed Informer long-term prediction model to obtain the best prediction model; Step S104: Based on the best prediction model, perform over-limit prediction on the gas concentration T1 of the working face, and comprehensively evaluate the prediction results based on the pre-determined accuracy evaluation index.

2. The adaptive long-time prediction method for gas overrun according to claim 1, characterized in that In step S101, Based on the Holt exponential smoothing algorithm, perform missing value processing on the obtained multi-source data in the coal mine underground; Standardize the multi-source data in the coal mine underground after missing value processing, and identify outliers in the standardized data based on the 3σ method; Based on the autoregressive model, process the outliers identified in the multi-source data in the coal mine underground; For the multi-source data in the coal mine underground after outlier processing, construct the maximum value Y of 5 min, 30 min, 1 h, and 24 h respectively t,max , the minimum value Y t,min , the average value Y t,ave , the difference value Y t,dif .

3. The adaptive long-time prediction method for gas overrun according to claim 2, wherein, According to the autoregressive model: Process the outliers identified in the multi-source data in the coal mine underground; where M is the outlier at time i in the multi-source data in the coal mine underground; c is a constant term; p is the autoregressive order, representing the time interval; is the autocorrelation coefficient; ε is the random error value; X i-j is the historical observation value at time j before time i in the multi-source data.

4. An adaptive long-time prediction method for gas overrun according to claim 2, characterized in that According to the formula: For the multi-source data in the coal mine underground after outlier processing, extract the maximum value Y at 5 minutes, 30 minutes, 1 hour, and 24 hours respectively t,max , the minimum value Y t,min , the average value Y t,ave , the difference value Y t,dif ; Among them, t represents different time scales; x t represents the multi-source data in the coal mine underground after outlier processing within the time range of t; x t-1 represents the multi-source data in the coal mine underground after outlier processing within the previous time range of t.

5. An adaptive long-term prediction method for gas overrun according to claim 1, characterized in that In step S102, it includes: Calculate the Pearson correlation coefficient between the gas concentration T1 of the working face and the different time-scale features, and adaptively extract the gas over-limit prediction indexes with the absolute value of the Pearson correlation coefficient in [0.4, 1] to form the best prediction index combination.

6. The self-adaptive long-time prediction method for gas overrun according to claim 1, wherein In step S103, Determine that the number of layers of the multi-head probability sparse self-attention mechanism, the number of layers of self-attention distillation, and the number of fully connected layers of the Informer long-term prediction model are the optimization parameter combination of the Informer long-term prediction model.

7. An adaptive long-time prediction method for gas overrun according to claim 1, characterized in that In step S103, the Informer long-term prediction model includes: an encoder and a decoder; The encoder is composed of 3 identical layers stacked, and each layer contains a multi-head probability sparse self-attention mechanism sub-layer and a self-attention distillation sub-layer; The decoder is composed of 2 identical layers stacked, and each layer contains a masked multi-head probability sparse self-attention mechanism sub-layer and a self-attention mechanism sub-layer.

8. An adaptive long-time prediction method for gas overrun according to claim 1, characterized in that In step S104, according to the formula: Calculate the mean absolute error MAE, mean square error MSE, root mean square error RMSE, and coefficient of determination R between the prediction result obtained by the optimal prediction model and the true value of the gas concentration of the target working face 2 ; where H k is the k-th true value of the gas concentration of the working face; h k is the prediction result corresponding to the k-th true value of the gas concentration; J k is the mean value of the K true values of the gas concentration of the working face; k ∈ K, and both k and K are positive integers.

9. An adaptive long-time prediction system for gas overrun, characterized in that Including: The source data unit is configured to process the acquired multi-source data {T0, T1, T2, T7, FS, W, GT, GL, GW, GP} in the coal mine underground to construct different time-scale features of the multi-source data in the coal mine underground; where T0 is the gas concentration at the upper corner of the working face, T1 is the gas concentration of the working face, T2 is the gas concentration in the return airway, T7 is the gas concentration in the refuge chamber, FS is the full working face wind speed, W is the working face temperature, GT is the gas drainage concentration in the goaf, GL is the gas drainage flow rate in the goaf, GW is the gas drainage temperature in the goaf, and GP is the gas drainage pressure in the goaf; The index unit is configured to adaptively screen the best prediction index combination of the working face gas concentration according to the Pearson correlation coefficient of the different time-scale features of the working face gas concentration; The parameter optimization unit is configured to optimize the optimized parameter combination of the pre-constructed Informer long-term prediction model based on the gray wolf optimization algorithm to obtain the best prediction model; The comprehensive evaluation unit is configured to perform over-limit prediction on the working face gas concentration based on the best prediction model, and comprehensively evaluate the prediction results based on the pre-determined accuracy evaluation index.

Images