Study on the risk identification of abnormal gas outbursts based on the mechanism of biological immunity

Abstract

With the depletion of shallow coal resources, mining is gradually extending to deeper levels, with increasing gas content and gas pressure. Aabnormal gas outbursts phenomena occurs from time to time, posing a serious threat to coal mine safety production. Therefore, the risk identification of abnormal gas outbursts is of urgent practical significance. However, the process of identifying abnormal gas outbursts risk imposes higher requirements on the ability of adaptability, self-learning, and self-organization. The artificial immune technology based on the biological immune mechanism provides a powerful information processing and problem-solving paradigm, showing certain advantages in dynamic risk identification, which can meet the needs of dynamic risk identification of abnormal gas outbursts. In order to achieve dynamic risk identification of abnormal gas outbursts under complex underground conditions, provide decision-making basis for rapid warning and early prevention of abnormal gas outbursts, the principles of immune recognition of biological immune systems were adopted. Research was conducted on the identification of spatial risk zones for abnormal gas outbursts, the adaptive recognition algorithm based on T-B cell principles, and the type recognition of abnormal gas outbursts based on the Dynamic Time Warping algorithm. A risk identification model for abnormal gas outbursts based on the biological immune mechanism was constructed and tested in the 9111 mining face of a mine in Huaibei. The results show: (1) The adaptive recognition algorithm based on T-B cell principles can adaptively recognize the characteristic vectors of abnormal gas outbursts by adaptive adjustment of detectors and cloning and mutation of learning vectors, achieving the recognition and memorization of known or unknown feature vectors under dynamically changing environmental conditions. (2) The adaptive recognition algorithm for abnormal gas outbursts based on T-B cell principles and the type recognition algorithm for abnormal gas outbursts based on the Dynamic Time Warping algorithm, combined with the characteristics of biological immune systems, construct a risk identification model for abnormal gas outbursts based on the biological immune mechanism, which has the characteristics of adaptability, learning, and memorization. (3) Taking a abnormal gas outburst event in a certain 9111 working face of a mine in Huaibei as an example, the model was verified by inputting the characteristic vectors of abnormal gas outbursts and the output of the risk identification of abnormal gas outbursts based on the biological immune mechanism. The research results show that the dynamic risk identification model for abnormal gas outbursts based on the biological immune mechanism can meet the requirements of problem-solving in constantly changing complex environments, achieve dynamic risk identification of abnormal gas outbursts, and provide a basis for risk warning and intelligent decision-making for abnormal gas outbursts in mines.

Introduction

Coal is the foundation energy and main energy source in China, especially in the past 20 years of the 21st century¹, the coal production has increased from 11.06 billion tons in 2001 to 40.7 billion tons in 2021, almost doubling². The sharp increase in coal demand has led to the depletion of shallow resources, forcing coal mining to move to deeper areas, with kilometer-deep mines becoming more common^3,4,5. Parameters such as ground stress, gas pressure, gas content, etc., also increase with the depth of mining, leading to the occurrence of abnormal gas outbursts during the coal mining process. abnormal gas outbursts are the main factor triggering gas accidents^6,7,8,9. Coal mines typically have a normal gas outburst situation underground, but when factors such as mining activities, ground stress, and atmospheric pressure affect the coal seams, abnormal gas outbursts can be triggered, leading to a sudden increase in gas outbursts within a short period of time. This can easily cause the underground gas concentration to exceed limits, posing risks such as gas explosions and gas combustion if effective control measures are not taken. Failure to effectively manage excessive gas can also result in asphyxiation incidents. For example, on April 10, 2019, a major accident occurred at the Liziyi coal mine in Sichuan province, where 3 people died due to a coal wall collapse accompanied by abnormal gas outbursts. On May 28, 2019, a gas outburst accident occurred at the Xinglong coal mine in Hunan province, resulting in 5 deaths and 1 injury. On July 31, 2019, a gas explosion occurred at the Dashu coal mine in Guizhou province due to abnormal gas outbursts in the old goaf area, causing 7 deaths and 1 injury. Therefore, the risk control situation of coal mine abnormal gas outbursts remains severe, and strengthening the risk control of abnormal gas outbursts is particularly important.

The primary task of preventing and controlling the risk of abnormal gas outbursts is to effectively identify the risk of abnormal gas outbursts. Some scholars have identified the abnormal gas outburststate in the working face using traditional mathematical statistical models and methods. Han Songlin¹⁰ applied theories such as gas geology and mathematical statistics, based on the division of gas geological zones, combined with the occurrence of gas in Shunhe coal mine, analyzed the gas geological control effect of the No. 2 coal seam, discussed the geological influencing factors of abnormal gas outbursts, and established a prediction model for gas abnormal areas, verifying the effectiveness of the model. Yang Yanguo¹¹ used the Shapiro-Wilk and Lilliefors joint normality test method to explore the distribution characteristics of gas concentration time series, and realized the identification and real-time early warning of abnormal gas outbursts in the working face. Wei Lianjiang¹² applied the K-line chart theory to identify the abnormal gas concentration sequences of different mines, and the accuracy of the identification results was relatively high. Based on anomaly identification, Ma Lijuan¹³ used fuzzy mathematics principles and pattern recognition technology to establish a danger identification and prediction model based on the rate of change of gas concentration, dynamically predicting the degree of danger in different areas of the working face. Moreover, many researchers have conducted research on the identification of gas emission risk using machine learning algorithms. Chen Zuyun^14,15 used support vector machine algorithms to identify the gas emission risk information in roadway excavation faces, obtaining the feature vector of monitoring gas concentration in roadway excavation faces. Wang Feiyin¹⁶ established an image recognition model based on Relief F-dimensional interval-support vector machine by analyzing the characteristics of gas concentration changes in images after abnormal gas outbursts in roadway excavation faces to identify the degree of risk of gas abnormal eruption according to the changes in gas images. Yang Shouguo¹⁷ and others not only trained neural network models for identifying abnormal gas outbursts types, but also developed a set of software that can quickly identify abnormal gas outbursts types and locations using a multi-sensor joint decision-making method, successfully verifying the practicality of the software. Risk identification is the prerequisite for accident prevention. Only by identifying the risks can corresponding measures be taken for prevention and control. Yang Bo¹⁸ et al. proposed a new dynamic risk recognition model based on the advantages of support vector machine and artificial immune algorithm in risk dynamic recognition. Firstly, the original data was subjected to feature selection and dimensionality reduction processing. Then, the penalty parameters and kernel function parameters of support vector machine (SVM) were selected through artificial immune optimization algorithm (IOA). Combining the advantages of multi classification methods of support vector machine, a dynamic risk recognition model based on the integration of support vector machine and immune optimization algorithm was proposed. The experimental results on the Heart Disase dataset showed that the forward and reverse antigen recognition rates of the model were 95.82% and 96.01%, respectively, both higher than traditional recognition models. Li Min¹⁹ et al. aimed to quantify the risk of coal mine gas explosions and address the lack of uncertainty in risk analysis. Based on expert experience, they identified the main risk factors affecting gas explosions, constructed a risk topology model, and evaluated the prior and conditional probabilities of risk factors based on fuzzy set theory. Then use Bayesian forward causal reasoning to calculate the probability of gas explosion occurrence. Finally, sensitivity analysis was completed using Bayesian importance analysis. A new method for gas explosion risk assessment has been proposed. Zhang Jufeng²⁰ et al. used dynamic data-driven technology to build a abnormal gas outbursts risk warning system architecture to solve the problem of excessive reliance on system models in the process of coal mine abnormal gas outbursts risk warning, inability to dynamically and real-time correct prediction models, and low accuracy and reliability of prediction and forecasting. They explored key technologies such as dynamic data-driven gas outburst monitoring curve fitting, dynamic warning model selection and correction, and warning system development, and developed a abnormal gas outbursts risk warning system software based on dynamic data-driven technology.Whether traditional mathematical statistical models, wavelet analysis, fuzzy mathematics pattern recognition methods, or the current machine learning methods that scholars are studying more, are mostly static models, with few studies involving the dynamic identification of abnormal gas outbursts. Therefore, further exploration is needed for dynamic risk identification methods of abnormal gas outbursts.

With the rapid development of artificial intelligence, new ideas and methods have been provided for the identification of abnormal gas outbursts risks in coal mines. Artificial immune systems, which draw on the immune mechanisms of biology, are one of the frontiers in artificial intelligence research. They possess characteristics such as dynamism, adaptability, distribution, and robustness. They have been widely applied in fault diagnosis, information processing, machine learning, intelligent optimization, and other fields, reducing reliance on experience to a certain extent and demonstrating more autonomy. Given the complex and changing environment of gas outburst in the coal production process, abnormal gas outbursts exhibit uncertainty and spatiotemporal coupling. Similar to the immune system, where and when antigens invade the system are unknown, the immune system can accurately identify known or unknown antigens, produce corresponding antibodies to eliminate antigens, and maintain stability in a changing environment. This is consistent with the requirement to effectively identify and prevent risks of abnormal gas outbursts in the dynamically changing underground environment. Based on the analysis of the immune recognition mechanism of the biological immune system, this study establishes a risk identification index system for abnormal gas outbursts, constructs a abnormal gas outbursts risk identification model based on immune recognition mechanism, and applies it in the 9111 working face of a mine in Huaibei, verifying the immune recognition ability of abnormal gas outbursts.

Basic principles of biological immune recognition

The immune system is a pattern recognition system, mainly completed by macrophages and lymphocytes. Among them, macrophages have a universal recognition of antigens, relying on their powerful phagocytic ability to engulf and process all antigens that invade the body. If an antigen cannot be eliminated, the antigenic determinant cluster will be exposed, activating the specific immune recognition of the immune system; the specific recognition is carried out by lymphocytes, mainly relying on the specific binding of T lymphocyte and B lymphocyte receptors on the membrane surface with the antigenic determinant cluster to recognize antigens^21,22,23. If the antigenic determinant base of a certain antigen is different from the receptor structure, they cannot bind, and lymphocytes cannot recognize the antigen and proceed to the next immune process; if the antigenic determinant base of a certain antigen is similar to the receptor structure, they will bind and recognize the type of antigen. According to the type of antigen, lymphocytes generate antibodies corresponding to that antigen through a series of proliferation and differentiation processes to clear the antigen²⁴. The affinity between the antigenic determinant cluster and the lymphocyte antigen receptor is measured by affinity, with a higher similarity in structure leading to a higher affinity. When the affinity between the two exceeds a certain threshold, lymphocytes can recognize that specific antigen^25,26,27. The principles of lymphocyte immune recognition and the schematic diagram of the immune recognition process of the immune system are shown in Figs. 1 and 2.

Fig. 1

Schematic diagram of the immune recognition function of lymphocytes.

Fig. 2

Schematic diagram of the immune system immune recognition process.

Immune recognition is the first step to achieve an immune response, belonging to the induction phase of the immune response. When an antigen first invades the body, immune cells such as macrophages, B lymphocytes, and T lymphocytes will process and recognize the antigen. Based on the antigen recognition results, B lymphocytes and T lymphocytes will further proliferate and differentiate, thereby generating effector cells and memory cells, to achieve antigen clearance by producing specific antibodies; when the antigen invades the body again, memory cells will rapidly proliferate and differentiate, producing more antibodies against the antigen to eliminate it more quickly²⁸. Therefore, the immune recognition of biological immune system has the characteristics of learning, adaptability, and memory.

The immune system mainly extracts antigen features through macrophages, presents the feature to T cells for judgment, generates effector molecules based on the judgment results, activates B cells for proliferation and differentiation, further produces effector cells and memory cells against the antigen. Effector cells eliminate the antigen by producing specific antibodies, while memory cells are preserved by the biological immune system to form memory against the antigen^29,30. Through a series of processes such as feature extraction, antigen recognition, antigen memory, and antigen elimination, the immune system can effectively clear the antigen, maintaining the stability of the body.

The gas emission environment faced by coal production space is uncertain, and the form of gas emission (including normal emission and abnormal emission) is accidental, and when and where it invades the system is unknown in advance. This is very similar to the biological immune system, which can recognize and defend against both known and unknown antigens. This fully conforms to the characteristics of gas dynamic risk identification and prevention. Therefore, simulating the operation mechanism of the biological immune system and constructing a dynamic risk identification model for abnormal gas outbursts based on immune mechanism is very important for preventing abnormal gas outbursts.

Abnormal gas outbursts risk identification indexes

Characteristics indicators of abnormal gas outbursts

From the perspective of the relationship between gas concentration, factors affecting abnormal gas outbursts, outburst quantity, and prevention and control measures, the factors affecting abnormal gas outbursts can lead to abnormal changes in gas outburst quantity. Prevention and control measures (such as ventilation, gas extraction) can reduce or eliminate the impact of abnormal gas outbursts. The risk of abnormal gas outbursts is the result of the interaction between the two. If the influence of inducement factors is greater than the influence of prevention and control measures, the risk of abnormal gas outbursts increases; if the influence of prevention and control measures is greater than the influence of inducing factors, the likelihood of abnormal gas outbursts decreases and the severity of consequences is reduced. Gas concentration reflects the interaction between gas outburst quantity and air volume (prevention and control measures), and gas concentration does not require the consumption of a large amount of manpower and material resources like drilling cuttings, gas liberation index, and other indicators. It is relatively easy to obtain, can be recorded in real time, and sensitively reflects changes in gas outburst quantity. The relationship between gas concentration, outburst quantity, factors affecting abnormal gas outbursts, and prevention and control measures is shown in Fig. 3.

Fig. 3

Schematic diagram of the relationship between gas concentration, air volume, influencing factors, and prevention and control measures.

The gas concentration index has the characteristics of easy data acquisition, real-time, sensitivity, objectivity, etc., so it is reasonable and feasible to use the gas concentration as a risk feature index for abnormal gas outbursts in a certain working face for the risk identification of abnormal gas outbursts .

Feature vectors

Real-time gas concentration can reflect the state of gas outburst at a certain moment and location. The higher the real-time gas concentration, the higher the possibility of gas exceeding limits, the higher the abnormality of gas outburst, and the greater the danger. By consulting relevant literature, indicators such as moving average, standard deviation, volatility, coefficient of variation, ratio of variation frequency, and relative rate of change can be used to analyze the time series of gas concentration and understand the changes in gas concentration³¹.

Moving average μ

The moving average can dynamically reflect the state of gas outburst and its trend (increase, unchanged, decrease) over a period of time. The moving average can be represented as Eq. (1).

$${\mu _t}=\left\{ \begin{gathered} \frac{1}{k}\mathop \sum \limits_{{i=1}}^{k} {C_{t - i+1}}~~~~\left( {t \geqslant k} \right) \hfill \\ \frac{1}{t}\mathop \sum \limits_{{i=1}}^{t} {C_i}~~~~~~~~~~\left( {t<k} \right)~~ \hfill \\ \end{gathered} \right.~$$

(1)

In the formula:

μ_t——the moving average of the gas concentration data of the t-th item newly monitored sequence.

k——the fixed number of data items for gas concentration time series with fixed gas concentration.

C_i——the i-th real-time gas concentration data.

Among them, there may exist cases where t < k in the original sequence, so the moving average of the t-th item gas data is taken as the average of the actual number of gas data items.

Standard deviation Σ

Standard deviation can reflect the degree of dispersion of each sequence. Like moving averages, it also has the characteristic of being “mobile”, constantly calculating the standard deviation of new sequences as gas data is updated. The larger the standard deviation, the greater the degree of dispersion of the sequence, and the higher the possibility of abnormal gas outburstsurging out. As per formula (2), the standard deviation can be expressed as:

$${\sigma _t}{\text{=}}\left\{ \begin{gathered} \sqrt {\frac{1}{k}\mathop \sum \nolimits_{{i=1}}^{k} {{\left( {{C_{t - i+1}} - {\mu _t}} \right)}^2}} ~~~~~~~~~\left( {t \geqslant k} \right) \hfill \\ \sqrt {\frac{1}{t}\mathop \sum \nolimits_{{i=1}}^{t} {{\left( {{C_i} - {\mu _t}} \right)}^2}} ~~~~~~~~~~~~~~\left( {t<k} \right) \hfill \\ \end{gathered} \right.$$

(2)

In the formula:

σ_t——the standard deviation of the gas concentration data of the t-th item being monitored.

μ_t ——the moving average of the sequence in which the new monitored gas concentration data at time t is located;

k——the fixed number of data items for gas concentration time series with fixed gas concentration.

C_i——the real-time gas concentration data of the i-th item.

Volatility B

Volatility is one of the most sensitive indicators reflecting the abnormal changes in gas outburst, and it can also reflect the activities of coal and rock mass. Volatility can measure the degree of changes in gas concentration over a certain period of time. Generally, the greater the volatility or the more frequent the positive or negative changes, the higher the degree of gas abnormality. According to formula (3), volatility can be expressed as:

$${B_t}=100\frac{{{C_t} - {\mu _{t - 1}}}}{{{\mu _{t - 1}}}}$$

(3)

In the formula:

B_t——the volatility of the gas data for the t-th new monitoring;

C_t——the real-time gas concentration data for the t-th term;

μ_t-1——the moving average of the gas concentration data for the (t-1)th item newly monitored sequence.

Coefficient of variation V

The coefficient of variation is the ratio of the standard deviation of a gas data sequence to its moving average. Different from the standard deviation, the coefficient of variation takes into account both the standard deviation and the mean, and has an advantage in comparing the dispersion of more than two groups of sequences. The larger the coefficient of variation, the greater the dispersion of the gas data and the higher the possibility of abnormal gas outbursts. The coefficient of variation can be represented by Eq. (4).

$${V_t}=\frac{{{\sigma _t}}}{{{\mu _t}}}$$

(4)

In the formula:

σ_t——the standard deviation of the gas concentration data of the t-th item being monitored.

μ_t——the moving average of the sequence in which the new monitored gas concentration data at time t is located;

Ratio of variance frequency P

The variable frequency ratio is the ratio of the frequency of gas concentration changes in the current sequence to the average frequency of gas concentration changes in the previous three sequences, which can reflect the stability of gas emission. If the frequency ratio of gas data variation gradually increases over time, it indicates a higher possibility of abnormal gas outbursts. In order to better understand the frequency of changes in the current cycle gas data, the following mathematical expression is used to explain it³²: when forming a gas concentration time series {C_t−k+1,C_t−k+2,.,C_t−1,C_t}, label the time series in order to generate $\{ C_{1}^{\prime },C_{2}^{\prime },.,C_{{x - 1}}^{\prime },C_{x}^{\prime }\}$. If$\Delta C_{x}^{\prime }=C_{x}^{\prime } - C_{{x - 1}}^{\prime } \ne 0$, then the frequency of changes in the series is n_x=n_x−1+1, where x$\in$(1,2,3.,k),$C_{0}^{\prime }$is the first gas concentration data of the previous series, n₀ = 0. According to formula (5), the variable frequency ratio can be expressed as:

$${P_t}{\text{=}}\frac{{{F_t}}}{{\frac{1}{3}({F_{t - 1}}+{F_{t - 2}}+{F_{t - 3}})}}$$

(5)

In the formula:

P_t —— the frequency ratio of changes in the sequence where the newly monitored gas concentration data of the t-th item is located.

F_t——the number of frequency of the value changes in the sequence of the newly monitored t-th gas concentration data.

The relative change rate v

The relative rate of change reflects the degree of change of gas monitoring data compared to the previous monitoring data. The greater the relative rate of change, the greater the degree of change in gas monitoring data, and the higher the possibility of gas outburst anomaly. The relative rate of change can be expressed by formula (6).

$${v_t}=\frac{{{C_i} - {C_{i - 1}}}}{{\Delta t}}$$

(6)

Through the above analysis, compared to real-time gas concentration, the moving average considers the overall situation of the sequence, which has a certain lag in identifying the risk of abnormal gas outbursts; the coefficient of variation has a greater advantage in the analysis of the dispersion degree of multiple sequence compared to standard deviation; the relative rate of change can reflect the changes in gas concentration in a short period of time. Therefore, this paper selects real-time gas concentration, coefficient of variation, volatility, and relative rate of change as the characteristic indicators of gas concentration change. Among them, real-time gas concentration can reflect the real-time status of gas concentration, belonging to the state indicator; volatility, coefficient of variation, and relative rate of change can reflect the fluctuation and trend of gas concentration, belonging to the trend indicator. By using these four indicators, the state and trend of gas concentration can be comprehensively analyzed, thereby identifying the state of gas outburst and the risk of abnormal gas outbursts. Gas concentration data can be obtained online in real time, so each indicator can provide a basis for the real-time and dynamic identification of gas outburst status and abnormal gas outbursts risk.

To reduce data dimensionality and improve the computational speed of subsequent recognition algorithms, the volatility, coefficient of variation, and relative rate of change are integrated using the importance of various indicator features to determine a comprehensive discrimination index, defined as the Gas Variation Index (GVI) and denoted as P. The calculation process is as follows:

Assume there is a gas concentration sequence T={C₁, C₂, C₃, …, C_n} with n items, utilizing the XGBoost algorithm to obtain weights of volatility, coefficient of variation, and relative rate of change as w₁, w₂, w₃. Based on the idea of dynamic sliding window, calculate the volatility set B={B₁, B₂, B₃, …, B_n}, coefficient of variation set V={V₁, V₂, V₃, …, V_n}, and relative rate of change set v={v₁, v₂, v₃, …, v_n} for the gas concentration sequence. Then, the Gas variation index P_n for the nth item is:

$${P_n}={w_1} \times {B_n}+{w_2} \times {V_n}+{w_3} \times {v_n}$$

(7)

By the above calculations, the collection of gas change indices for gas concentration sequence forms a set P={P₁, P₂, P₃ ,…, P_n}. Combining each gas concentration data with its corresponding Gas variation index forms a feature vector of n abnormal gas outbursts cases, i.e., (T, P)={(C_t, P_t), t = 1, 2, 3, …, n}.

Classification of abnormal gas outbursts risk levels

To reasonably divide the level of abnormal gas outbursts, based on the selection of abnormal gas outbursts characteristic indicators and the generation of characteristic vectors, the gas concentration index and gas variation index in the abnormal gas outbursts characteristic vectors are divided into high, relatively high, relatively low, and low levels, corresponding to the anomaly level I, II, III, IV. The levels of abnormal gas outbursts and their definitions are shown in Table 1.

Table 1 Gas outburst anomaly levels and their definitions.

Abnormal gas outbursts risk immune identification of antigen tissue

Morpho-spatial model for immunity identification of abnormal gas outbursts risk

The abnormal gas outbursts characteristic vector forms the antigen determinant base of gas outburst state and abnormality degree. To visually represent the process of identifying the abnormality degree of abnormal gas outbursts, a morphological space model for recognizing the abnormal degree of abnormal gas outbursts has been constructed. The morphological space model is a classic model for studying artificial immune systems and is one of the important methods for describing immune events such as immune recognition and immune learning. In the immune recognition process, antigen recognition and elimination are mainly achieved through the binding of antigens with lymphocytes or antibodies. The higher the degree of binding, the stronger the recognition ability of lymphocytes and antibodies. Similarly, based on describing the distribution of antigens and antibodies in the morphological space, the morphological space model uses affinity to measure the degree of binding between the two, thereby studying the process of immune recognition.

As shown in Fig. 4, define an L-dimensional shape space V, in which there are a certain number of antibodies (represented by ·) and antigens (represented by ×). Suppose the monitoring radius of an antibody (i.e., an immune detector) is r. If an antigen is within the monitoring range of an antibody, it means the antigen is recognized by the antibody; otherwise, if the antigen is outside the monitoring range of the antibody, the antigen cannot be recognized by the antibody.

Fig. 4

Schematic diagram of morphological space.

The space where gas anomalous ejection abnormal feature vector distribution is termed as the gas anomalous ejection abnormal degree recognition space. In the gas anomalous ejection abnormal degree recognition space, the feature vectors are regarded as antigens, and the detectors as antibodies, judging whether antigens and antibodies can match by calculating the affinity between antigens and antibodies, i.e., if the affinity between the two is greater than the matching threshold, the antigen can be recognized.

In the recognition space of anomaly severity, the affinity between antibodies (detectors) and antigens (feature vectors) is:

$$aff\left( {X,Y} \right)=\frac{1}{{1+{d_E}}}$$

(8)

The d_E (Euclidean distance) is the distance between the feature vector of abnormal gas outbursts and the detector in the recognition space. The calculation formula of Euclidean Distance is:

$${d_E}=\sqrt {\mathop \sum \limits_{{i=1}}^{L} {{\left( {a{b_i} - a{g_i}} \right)}^2}}$$

(9)

When the distance between antigen ag_i and antibody ab_i is less than a certain matching threshold, the degree of abnormal gas outbursts anomaly is identified. The smaller the Euclidean distance between an antigen and a specific antibody, the more similar the degree of abnormal gas outbursts anomaly recognized by that detector.

Identification of Spatial risk areas for abnormal gas outbursts

In the space of abnormal gas outbursts anomaly recognition, the recognition space is divided into four regions based on the anomaly degree classification criteria, represented by M₁, M₂, M₃, M₄ for the regions of abnormal gas outbursts anomaly degrees I, II, III, and IV in the recognition space V. The anomaly degree recognition detectors in the four spatial regions are respectively the first-level detector, the second-level detector, the third-level detector, and the fourth-level detector. The division of the boundaries of the M₁, M₂, M₃, M₄ four anomaly degree level regions should be determined based on the division criteria of two indicators in the feature vector, and the correspondence between the feature vector and the anomaly degree of abnormal gas outbursts is not fixed, and can be flexibly adjusted according to the actual production of different coal mines.

Given that the abnormality recognition space of gas outburst belongs to a two-dimensional space, the space is divided by using ellipses as boundaries. The abnormality levels within three ellipse regions are IV, III, and II, while the areas outside the three ellipses are level I. The process of abnormal gas outbursts recognition starts by traversing the detector in the IV level region, then traversing the regions in the III, II, and I levels, ultimately achieving the traversal of the entire abnormal gas outbursts recognition space detector. In order to achieve full coverage of the abnormal gas outbursts recognition space, a correction coefficient of the major and minor axes of the ellipse boundaries is introduced to optimize the coverage area of each region, such that the lengths of the major and minor axes of the ellipse become G times the original ellipse axis length, thereby appropriately adjusting the coverage area of each region’s ellipse. As shown in Fig. 5, by analyzing the division criteria of the feature vector indicators, the correction coefficient G of the three ellipses in the recognition space is determined, optimizing the division of the abnormality recognition space areas.

Fig. 5

Schematic diagram of dividing recognition space regions by elliptical boundaries with correction factors.

Assuming the abnormal gas outbursts identifies the spatial boundaries of the ellipse, the major axis is in the x-axis direction and the minor axis is in the y-axis direction. The high, relatively high, relatively low, and low levels of abnormal gas outbursts feature vector antigen indicators ag_x and ag_y points are x_n, y_n (where n = 1, 2, 3) respectively. Based on determining the indicator points, the centers of the ellipses are determined to be (x_n/2, y_n/2), and a correction coefficient G is added to optimize the ellipses. The lengths of the major and minor axes of the ellipses become 1/2G_nx_n and 1/2G_ny_n. Therefore, the equation of the abnormal gas outbursts identification spatial boundary ellipse is determined as $\frac{{{{(x - \frac{{{x_n}}}{2})}^2}}}{{{{(\frac{{\text{1}}}{{\text{2}}}{G_n}{x_n})}^2}}}+\frac{{{{(y - \frac{{{y_n}}}{2})}^2}}}{{{{(\frac{{\text{1}}}{{\text{2}}}{G_n}{y_n})}^2}}}=1$. Based on the division of the abnormal gas outbursts identification spatial area, the detectors in the identification space can determine the level of abnormality by comparing the polar coordinates of the detector’s center with the three boundary ellipses.

By dividing the space for identifying the degree of abnormal gas outbursts through ellipses, and ensuring the rationality of the division, adjusting the different regions appropriately by correction coefficients, the identification of dividing areas can better reflect the reality, providing certain standards for recognizing the degree of abnormal gas outbursts .

Adaptive recognition algorithm based on T-B cell principle

Algorithm principles

The specificity recognition of the immune system mainly relies on the coordinated action of T cells and B cells. T cells gradually develop and mature in the thymus based on the principle of negative selection, primarily responsible for recognizing antigens by receiving antigen information substances (antigen-MHC complexes) and activating the proliferation and differentiation of B cells through the production of effector molecules, regulating the immune process; B cells, activated by effector molecules, undergo proliferation and differentiation, produce corresponding antibodies to eliminate antigens, generate memory cells, and prepare for a faster and stronger secondary immune response. T cells mainly play a role in recognition, regulation, and initiation of immune responses, while B cells not only recognize antigens but also learn and remember antigens through cloning and variation. The adaptive recognition algorithm based on the principles of T-B cells is designed according to the principles of T and B lymphocytes. The algorithm mainly consists of T modules and B modules, where the T module is the discriminative module corresponding to the formation of T lymphocytes and their antigen recognition function, and the B module is the learning module corresponding to the process of B lymphocytes generating learning and memory through proliferation and differentiation. The T module can discriminate the state and degree of abnormal gas outbursts, identifying the degree of risk of abnormal gas outbursts; the B module can learn the risk feature vectors of abnormal gas outbursts that were not detected by the detector and produce learning encoding, optimize the T module based on the learning encoding to achieve adaptive recognition of abnormal gas outbursts risks.

Algorithm process

Given that the adaptive recognition algorithm based on the T-B cell principle has characteristics such as adaptability, learning ability, and memory, which can meet the demand for dynamically identifying the constantly changing gas venting state and its abnormality level, the algorithm’s T module and B module algorithm processes are introduced below.

Algorithm flow of module T

The T module belongs to the discriminative module, with its flowchart shown in Fig. 6.

Fig. 6

T module algorithm flowchart.

The specific process is described as follows:

Step 1: Determine the detector radius as r, the preset threshold value for the number of detectors as N, the Euclidean distance between detectors as d₁, the Euclidean distance between the abnormal gas outbursts feature vector and detectors as d₂, the set of mature detectors as P, and divide the morphological space into four regions M₁, M₂, M₃, M₄.

Step 2: Generate a candidate detector randomly in any area.

Step 3: Calculate the Euclidean distance d₁ between the candidate detector and all generated detectors. If the candidate detector has a Euclidean distance with any of the generated detectors, delete the candidate detector and return to Step 2; if the candidate detector has Euclidean distances with all generated detectors, add the candidate detector as a mature detector to P.

Step 4: When the number of mature detectors in P reaches the set threshold N, the initialization is complete; when the number of mature detectors in P does not reach the set threshold N, go back to Step 2.

Step 5: Input the abnormal gas outbursts feature vector in the form of real value coding.

Step 6: Calculate the Euclidean distance between the abnormal gas outbursts feature vector and all mature detectors in P. If the feature vector’s Euclidean distance is smaller than one detector, output the type of the region to which the detector belongs and end the identification process. If the feature vector’s Euclidean distance is larger than all detectors, activate module B.

Algorithm flow of module B

Module B belongs to the learning module of the algorithm, and its algorithm flowchart is shown in Fig. 7.

Fig. 7

B module algorithm flowchart.

The specific description of the algorithm is as follows:

Step 1: Set the initial learning vector quantity as L_a, clone learning vector as L_b, clone learning vector quantity upper limit as L_b−limit, clone learning vector death probability as f_die, the Euclidean distance between abnormal gas outbursts feature vector and learning vector as d₃, the Euclidean distance between learning code and detector as d₄, set A, B, C, D four regions near the abnormal gas outbursts feature vector.

Step 2: Generate an initial learning vector of quantity L_a randomly and distribute it randomly.

Step 3: Input the feature vector of abnormal gas outbursts that is not within the detection range of the detector.

Step 4: Calculate the Euclidean distance between the abnormal gas outbursts feature vector and the nearby learning vectors, and depending on the different distances between the feature vector and the learning vectors, different types of mutations occur in the learning vectors. For learning vectors between similar points in seque belonging to the cluster region of d₃>r_A or d₃>r_D , no mutation occurs; for learning vectors belonging to the range of r_rAB , inherent mutations occur; and for learning vectors belonging to the range of $r_{B} < d_{3} < r_{C}$, random mutations occur.

Step 5: Generate clusters through mutation, form learning encodings based on cluster centers, calculate the Euclidean distance d₄, between the learning encoding and all detectors, adjust detector positions if $d_{4} < r$, generate a new detector if $d_{4} > r$, where the radius of the detector is the minimum distance from the learning encoding to a nearby detector.

Step 6: When there is a abnormal gas outbursts characteristic vector input in the future, if the feature vector is in the vicinity of a certain learning code and the occurrence reaches a set threshold $\delta$, clone L_b learning vectors near the feature vector, and then ${L_b} \leqslant {L_{b - \lim it}}$, perform Step 4, Step 5.

Step 7: L_b clones of learning vectors die based on the death probability f_die, achieving the purpose of timely updating the learning vectors.

Step 8: Transfer the learning code generated through the mutation and clone learning process to module T, abnormal gas outbursts feature vector learning ends.

Identification of abnormal gas outbursts based on biological immune mechanisms

The immune system not only can distinguish between “self” cells and “non-self” antigens, but also can produce corresponding antibodies to effectively eliminate antigens, with the prerequisite of the immune system being able to recognize what type of antigen it belongs to. Similarly, identifying the factors leading to abnormal gas outbursts is an important prerequisite for formulating prevention and control measures. Therefore, based on identifying the gas outburst status and the degree of abnormality, it is also necessary to identify the types of abnormal gas outbursts, in order to provide a basis for the subsequent response measures. The identification of abnormal gas outbursts types generally starts with the change curve of gas concentration, recognizing the characteristics of gas concentration change curves under different influencing factors, and thus achieving the identification of abnormal gas outbursts types.

Abnormal gas outbursts type identification model based on dynamic time warping algorithm

To identify the type of a gas concentration time series, it is necessary to measure the similarity between the gas concentration time series in question and known category gas concentration time series, and then identify the type of the gas concentration time series. Considering that the gas concentration time series similarity measurement may involve non-equal lengths, Dynamic Time Warping^32,33 (DTW) algorithm searches for similar characteristics between the input template and the reference template, compresses, expands, or transforms certain vectors as needed to obtain the least twisted cost curve path for matching. This capability allows the DTW algorithm to accurately identify the time series features of different categories of abnormal gas outbursts in gas anomaly recognition, even if these features have certain scale or offset on the time axis. The situation of abnormal gas outbursts in coal mines is complex and variable, including various factors such as gas sensor calibration, faults, displacement, roof collapse, local ventilation stoppage, and coal and gas outbursts. The DTW algorithm can handle this complexity and uncertainty by calculating the shortest dynamic bent distance between the target sample and multiple weighted reference sequences, accurately determining the category of abnormal gas outbursts. The DTW algorithm is used as the method for measuring the similarity between gas concentration time series to discriminate the types of gas concentration time series. The DTW algorithm is a nonlinear time-warping pattern matching algorithm proposed by Itakura, with the main goal of finding the shortest distance between two time series. Based on dynamic programming principles, it matches similar points between the time series nonlinearly, and determines the similarity between time series based on the sum of distances of all similar points. A larger total distance indicates lower matching degree and smaller similarity between sequences, while a smaller total distance indicates higher matching degree and greater similarity between sequences. The schematic diagram of the DTW algorithm principle is shown in Fig. 8. The path used to calculate the distance between similar points is referred to as the warping path. The dynamic path warping illustration of the DTW algorithm is shown in Fig. 9.

Fig. 8

Illustration of the DTW algorithm principle.

Fig. 9

Dynamic time warping (DTW) algorithm dynamic path alignment diagram.

Assuming there are two time series Q={q₁, q₂, q₃, …, q_i, …, q_n} and S={s₁, s₂, s₃, …, s_j, …, s_m}, where the length of time series Q is n and the length of time series S is m, and both n and m are integers greater than 1, the distance between similar points in the sequences is denoted by d_{i, j}, and the calculation formula is as shown in Eq. (10):

$${d_{i,j}}=\left\| {{{\left( {{q_i} - {s_j}} \right)}^2}} \right\|$$

(10)

According to the above formula, calculate the distances between similar points in sequence, resulting in a distance matrix d of n rows and m columns:

$d = \left[ {\begin{array}{*{20}c} {d_{{1,1}} } & {d_{{1,2}} } & {...} & {d_{{1,m}} } \\ {d_{{2,1}} } & {d_{{2,2}} } & {...} & {d_{{2,m}} } \\ \vdots & \vdots & \ddots & \vdots \\ {d_{{n,1}} } & {d_{{n,2}} } & {...} & {d_{{n,m}} } \\ \end{array} } \right]$

According to the idea of dynamic programming, the distance matrix d can be transformed into an n-row m-column cumulative distance matrix D as defined in Eq. (10), where the element D_{i, j} in the cumulative distance matrix D can be represented as:

$${D_{i,j}}={d_{i,j}}+\hbox{min} ({d_{i - 1,j}},{d_{i,j - 1}},{d_{i - 1,j - 1}})$$

(11)

A regular path is represented by W=(w₁, w₂, w₃, …, w_k), and it must satisfy the following four constraints:

(1) Boundedness:$\hbox{max} (n,m) \leqslant k \leqslant n+m+1$.

(2) Boundary conditions:

w₁ = D_1,1=d_1,1.

w_k=D_{n, m}=d_{n, m}.

All regular paths must take the endpoints of the diagonal of the cumulative distance matrix D as the starting and ending elements.

(3) Continuity: All elements in the regularity must not be discontinuous, and must be connected to each other.

(4) Monotonicity: if $w_{{k - 1}} = (i',j')$and ${w_k}=(i,j)$ exist, $\left\{ {\begin{array}{*{20}c} {i' \le i \le i' + 1} \\ {j' \le j \le j' + 1} \\ \end{array} } \right.$must satisfy.

The regular path must be monotonically increasing, ensuring that there is no crossover phenomenon while obtaining each point in two time series.

Between two time series, there may be many regular paths that satisfy the above constraints, but according to the objective of the DTW algorithm, the regular path with the smallest total length must be selected as the optimal regular path, which can be represented as:

$${w_k}=\left\{ {\begin{array}{*{20}{c}} {\begin{array}{*{20}{l}} {{D_{1,1}}} \\ {\hbox{min} ({D_{i - 1,j}},{D_{i,j - 1}},{D_{i - 1,j - 1}})} \\ {{D_{n,m}}} \end{array}}&{\begin{array}{*{20}{l}} {i=1,j=1} \\ {1<i<n,1<j<m} \\ {i=n,j=m} \end{array}} \end{array}} \right.$$

(12)

Through the above process, the optimal dynamic time warping distance obtained will be used as the basis for assessing the similarity between two time series. The closer the value of the optimal dynamic time warping distance is to 0, the more similar the two time series are; conversely, the smaller the similarity.

Through the introduction of the types of abnormal gas outbursts and the principle of DTW algorithm, known categories of abnormal gas outbursts time series sample data are used as reference sequences, the gas concentration time series to be evaluated is used as the sequence to be recognized, and the DTW algorithm is used to calculate the optimal alignment distance between the sequence to be recognized and all reference sequences. When the distance value is less than a certain threshold, the category of the sequence to be recognized is obtained. The abnormal gas outbursts type identification model based on DTW algorithm is shown in Fig. 10.

Fig. 10

abnormal gas outbursts type recognition model based on DTW algorithm.

Construction of abnormal gas outbursts risk identification model based on immunological mechanisms

Building on the foundation of the abnormal gas outbursts recognition model based on immune algorithm and the abnormal gas outbursts type recognition model based on DTW algorithm, the two algorithm models are combined, and based on the characteristics and principles of immune recognition system, a abnormal gas outbursts risk recognition model based on immune mechanism is constructed, as shown in Fig. 11.

Fig. 11

abnormal gas outbursts risk identification model based on immune mechanism.

The model is able to identify the degree and type of abnormal gas outbursts, mainly including data collection and analysis, identification stage, formulation of countermeasures, and memory bank.

• (1) Data collection and processing. The gas monitoring system collects methane data transmitted by methane sensors, generates methane concentration time series, and after analysis and processing, forms abnormal gas outbursts feature vectors. Its function is similar to the antigen processing of phagocytic cells in the immune system. Therefore, in this model, the gas monitoring system is equivalent to phagocytic cells in the immune system.

• (2) Identification stage. This process is the core part of the model, aiming to identify the abnormal degree and type of gas outburst, providing the basis for the formulation of countermeasures. The adaptive identification algorithm based on the T-B cell principle and the DTW algorithm are the core algorithms for immune recognition, corresponding to the recognition of the degree and type of antigen damage by the immune system. The adaptive identification algorithm based on the T-B cell principle recognizes the abnormal gas outbursts feature vector extracted from the gas concentration time series, outputting the gas outburst state and its abnormal degree; when the adaptive identification algorithm based on the T-B cell principle outputs an abnormal gas outburst state, the gas concentration sequence at this time is used as the input sequence for the DTW algorithm. The DTW algorithm recognizes the gas concentration sequence to output the type of abnormal gas outbursts. Through the above process, the abnormal gas outbursts risk level and type are output as the identification results, providing the basis for the formulation of subsequent countermeasures.

• (3) Development of countermeasures. This process is equivalent to the immune system’s response to antigens, where managers or experts formulate corresponding countermeasures based on the risk level and type of abnormal gas outbursts, in order to prevent and control the abnormal gas outbursts.

• (4) Memory bank. This process is similar to the immune system generating immune memory to antigens invading the body. It stores various reference sequences of abnormal gas outbursts as the type features of gas outbursts, and also stores the identification results and corresponding countermeasures generated in the identification stage. When gas outbursts of the same type and risk level reappear, they can be identified more quickly to prevent and control gas outbursts.

In summary, drawing on the characteristics of immune recognition, immune memory, vaccine injection, etc. of the biological immune system, this model has certain adaptability, learnability, and memorability. It can achieve dynamic recognition of abnormal gas outbursts appearing in constantly changing environments. Moreover, due to the model’s learnability and memorability, as the number of model recognitions increases, the model’s recognition ability will gradually improve.

Example analysis

The experimental analysis of the abnormal gas outbursts risk identification model based on immune mechanism is carried out on the 9111 mining face of a coal mine in Huainan. During the mining period, the geological conditions, air supply volume, and gas extraction measures of the working face did not change and remained stable.

Data collection and processing

The sample data was taken from the T₁ gas concentration sensor of the 9111 mining face. The T₁ sensor is installed within the range of 5–10 m from the outlet of the working face in the return air duct. The collected gas monitoring data was selected, and on the basis of preprocessing the gas monitoring data, three months of gas monitoring data were selected as the research sample. The gas sensor of this working face collects data every 10 s.

Calculation of feature indicator weights

To highlight the changes in feature points and reduce computational complexity, the concept of dynamic sliding window is utilized by setting the window size to 10, generating a gas concentration sequence based on dynamic sliding window. The three feature indicators of sample data-volatility, coefficient of variation, and relative change rate are calculated using their respective formulas. The sequences of the three feature indicators are obtained. To ensure the reliability of the weight calculation results, the sequences of the three feature indicators are normalized and input into the XGBoost algorithm. The feature importance of volatility, coefficient of variation, and relative change rate are 0.3, 0.44, and 0.26 respectively. Therefore, the weight vector of the three feature indicators is:

$$\begin{gathered} w=[{w_B},{w_V},{w_{\text{v}}}] \\ =[{w_1},{w_2},{w_3}] \\ =[0.3,0.44,0.26] \\ \end{gathered}$$

(13)

Risk level classification

Under normal production conditions, the highest methane concentration of the 9111 mining face is 0.29%, with an average concentration of 0.19%. According to the actual situation of the 9111 working face and the relevant requirements of the “Coal Mine Safety Regulations”, the methane concentration index and the methane change index are classified into different levels³⁴, as shown in Tables 2 and 3. In accordance with the relevant requirements of the “Coal Mine Safety Regulations” and the standard that “excessive methane concentration is an accident,” the methane concentration in the working face should be controlled within 1%. However, the current methane monitoring system only issues an alarm when the methane concentration reaches 1% or a certain set value, neglecting the analysis of methane concentration changes. Therefore, identifying abnormal changes in methane concentration within 1% and detecting abnormal gas outbursts early can help formulate corresponding decisions better and faster.

Table 2 Division criteria for gas concentration indexes.

Table 3 Classification criteria for gas variation index.

Based on the classification of gas concentration index and gas variation index level, the abnormality level is determined by combining the classification criteria of the two characteristic indices, as shown in Table 4.

Table 4 Classification criteria for the degree of abnormal gas outbursts.

Analysis of abnormal gas outbursts anomaly degree recognition model based on immune algorithm

According to the division criteria for abnormal gas outbursts anomaly degree recognition spaces M₁, M₂, M₃, and M₄ based on gas concentration index and Gas variation index, they correspond to abnormal gas outbursts anomaly degree levels IV, III, II and I. It can be seen from the division criteria using two characteristic indicators that the centers of ellipses used for dividing the M₁, M₂, M₃, and M₄ regions are (0.15, 0.15), (0.25, 0.25), (0.4, 0.3) respectively. Setting ellipse correction coefficients G₁ = 1.2, G₂ = 1.1, the boundary ellipse equations for dividing and recognizing spaces M₁ and M₂ are$\frac{{{{(x - 0.15)}^2}}}{{{{0.18}^2}}}+\frac{{{{(y - 0.15)}^2}}}{{{{0.18}^2}}}=1$; the boundary ellipse equations for dividing and recognizing spaces M₂ and M₃ are$\frac{{{{(x - 0.25)}^2}}}{{{{0.275}^2}}}+\frac{{{{(y - 0.25)}^2}}}{{{{0.275}^2}}}=1$; the boundary ellipse equations for dividing and recognizing spaces M₃ and M₄ are $\frac{{{{(x - 0.25)}^2}}}{{{{0.275}^2}}}+\frac{{{{(y - 0.25)}^2}}}{{{{0.275}^2}}}=1$.

Based on the division of the spatial identification of abnormal gas outbursts, the identification radius of detectors needs to be set. The monitoring radius of detectors, whether too large or too small, will affect the efficiency and accuracy of the model identification. If the identification radius of detectors is set too small, it will be necessary to generate more detectors to ensure the accuracy of identification, but it will also result in the overall efficiency of the model being low, and more detectors will become invalid detectors. If the identification radius of detectors is set too large, it is highly likely to cause the detectors to cross multiple identification zones, leading to misjudgments in the identification of the abnormal degree of abnormal gas outbursts. It is determined through comprehensive analysis that the identification radius of detectors for identifying the abnormal degree of abnormal gas outbursts is 0.04. The other parameter settings in the adaptive recognition algorithm based on T-B cell principle are as follows:

The initial learning vector quantity L_a = 1000, the upper limit of clone vector quantity L_b−limit = 3000, the range of inherent variation in region B is$0.5r < dis < r$in the recognition space, the range of random variation in region C is $r < dis < 1.5r$, the detector quantity threshold N = 80, the threshold parameter setting for clustering K = 200, and the number of occurrences of feature vectors near the learning code threshold δ = 3.

To verify the recognition capability of the abnormal gas outbursts anomaly degree identification model based on immune algorithm, the gas monitoring data of a abnormal gas outbursts at working face 9111 is selected as the test data for abnormal gas outbursts anomaly degree identification. The gas concentration change diagram of the abnormal gas outbursts at working face 9111 is shown in Fig. 12a. Using the calculation formulas for volatility, coefficient of variation, and relative change rate to calculate the indicators reflecting the abnormal change of gas outburst, as shown in Fig. 12b. Combined with the weights of each indicator, the gas variation index P of the gas concentration sequence is calculated, thereby generating the feature vector of abnormal gas outbursts (T, P)={(C_t, P_t), t = 1, 2, 3, ., n}.

Fig. 12

Schematic diagram of changes in gas concentration and characteristic indicators during abnormal outburst at the 9111 working face. (a) Diagram of the variation of gas concentration when anomalous outbursts occur, (b) Diagram of the variation of characteristic indicators.

Submit the normalized feature vector into the adaptive recognition algorithm based on T-B cell principle, realizing adaptive recognition of the feature vector through the movement and addition of detectors. The distribution of detectors and learning vectors is shown in Fig. 13, where the green circles in (a) indicate the movement or addition of detectors, and (b) shows the distribution of detectors after adaptive adjustment.

Fig. 13

Detector and learning vector distribution diagram, (a) Detector movement and addition (b) Detector and learning vector distribution.

According to Table 4, based on the gas concentration index and gas variation index of the 9111 working face, the abnormal gas outbursts degree are shown in Fig. 14.

Fig. 14

Abnormal degree identification results schematic diagram.

From Fig. 14, it can be seen that with a gas monitoring interval of 10, the gas is in a normal outburst state from t = 0 to t = 430 and from t = 460 to t = 760, with an abnormality level of level IV from t = 430 to t = 460 and t = 760 to t = 790, the gas outburst shows an abnormal state, requiring attention to the gas outburst state until it returns to normal; from t = 790 to t = 820, the abnormality level of gas outburst further increases, reaching level III, and management personnel need to stay alert; from t = 820 to t = 830, the abnormal level of gas outburst rapidly increases to level I equiring analysis of the cause of the abnormality and timely implementation of preventive measures.

Based on the immune algorithm, the abnormal gas outbursts risk recognition model can dynamically and adaptively identify the gas outburst state and its abnormal degree. It achieves learning and memorizing the feature vectors through cloning and mutation, and can timely understand the changes in gas outbursts based on the model’s output recognition results. In order to further identify the types of abnormal gas outbursts and formulate targeted preventive measures, the abnormal gas concentration sequence needs to be input into the abnormal gas outbursts type identification model based on the DTW algorithm for type recognition of abnormal gas outbursts.

Identification model analysis of abnormal gas outbursts types

Given that identifying the types of abnormal gas outbursts is an important basis for formulating targeted preventive measures, after identifying the degree of abnormal gas outbursts based on level of gas outbursts, it is necessary to identify the types of abnormal gas outbursts. That is, to identify the factors leading to abnormal gas outbursts through the gas concentration sequences generated under the influence of different factors. Through on-site investigations and reviewing on-site management data, the main reasons for abnormal gas outbursts in the past production process at 9111 working faces were mainly the following: cutting coal too quickly, stoppage of ventilation, collapse of the roof, and abnormal fluctuations before coal and gas outbursts. As the gas concentration data is mainly collected by sensors and according to relevant regulations, gas sensors need to be regularly calibrated, during which the gas concentration data may also undergo abnormal changes. Therefore, the abnormal gas outbursts identification model will also identify its abnormal degree. To avoid misjudgment due to sensor calibration, the gas concentration sequence during sensor calibration needs to be identified as a separate type. In order to deal with the noise interference that the DTW algorithm may encounter when processing gas abnormality data, generally, the data is smoothed by calculating the moving average or using a sliding window. In this paper, a sliding window is used to smooth the data to reduce the interference of noise on the analysis results. To address the potential irregularity issues in the data, continuous exploration and improvement of the algorithm itself are carried out to adapt to more complex data processing requirements and improve the accuracy of identification. For the instance analysis of the abnormal gas outbursts identification model based on the DTW algorithm, combined with the actual production at the 9111 working face, the gas concentration sequences under five conditions of cutting coal too quickly, stoppage of ventilation, collapse of the roof, abnormal fluctuations before coal and gas outbursts, and sensor calibration are collected as reference sequences. The identified abnormal gas concentration sequence is treated as the test sequence, and the DTW algorithm is used to calculate the optimal alignment distance between the test sequence and each reference sequence in turn. The five reference sequences are denoted as Q₁, Q₂, Q₃, Q₄, Q₅, the test sequence is denoted as S, and the optimal alignment distances between the test sequence and the reference sequences are denoted as D(1), D(2), D(3), D(4), D(5), with a designated optimal alignment distance threshold r = 2. Figure 15a-e shows a schematic diagram of the dynamic alignment paths between the test sequence and each reference sequence.

Fig. 15

Diagram of the dynamic programming path between the sequence to be sequenced and the reference sequence.(a) S-1 (b) S-2 (c) S-3 (d) S-4 (e) S-5

From Fig. 16, it can be seen that the optimal alignment distance D(3) between the test sequence and the various reference sequences is the smallest optimal alignment distance, and D(3) = 0.92 < r = 2. Therefore, the type of the abnormal gas concentration sequence is identified as roof collapse. Based on the model’s output for the identification of abnormal gas outbursts types, management personnel can quickly make corresponding decisions, formulate preventive measures for abnormal gas outbursts caused by roof collapses, such as increasing ventilation to dilute gas in a timely manner, reducing the gas concentration to below 1%; promptly evacuating personnel to ensure their safety; strengthening gas extraction and ensuring the qualification rate of gas extraction to reduce the gas content in coal seams and prevent gas from the source; strengthening daily roof management, promptly investigating and eliminating hidden dangers.

Fig. 16

Illustration of the optimal alignment distance between the test sequence and the reference sequence.

With the continuous increase in the number of recognition sequences, the obtained reference sequence will also be continuously updated. It is possible that a type of reference sequence may contain multiple sequences. However, based on the “minimum distance” principle, the abnormal gas outbursts type recognition model based on the DTW algorithm is also applicable to this situation, and the model’s recognition ability gradually improves as the reference sequences increase.

The risk identification model of abnormal gas outbursts based on artificial immune system is an innovative coal mine safety monitoring technology. It combines the principles and mechanisms of biological immune system, providing a new perspective and method for identifying the risk of abnormal gas outbursts. This model can capture the dynamic changes of abnormal gas outbursts in real time, rapidly respond to the fluctuation of gas concentration, and achieve real-time early warning of risks. Furthermore, the adaptability and robustness of artificial immune system enable the model to automatically adjust identification strategies, adapting to the risks of abnormal gas outbursts under different environments and conditions. At the same time, the model has strong capabilities in handling noise and abnormal data, maintaining stable identification performance in complex and changeable coal mine environments. Based on the recognition mechanism of biological immune system, the identification model based on artificial immune system can accurately identify the risks of abnormal gas outbursts.

The risk identification model of anomal gas outbursts based on artificial immunity can be applied to the early warning system of abnormal gas outbursts, realizing real-time monitoring and warning of gas concentration. When the gas concentration exceeds the safety threshold, the system will automatically issue a warning signal, reminding coal mine management personnel to take measures in a timely manner. Secondly, the model can also guide the optimization of the ventilation system by adjusting the operating parameters of ventilation equipment and the layout of ventilation networks through real-time monitoring of gas concentration and distribution, ensuring the air circulation and gas concentration control within a safe range in the coal mine. In the event of a gas accident, the artificial immune-based identification model can quickly identify the cause and scope of the accident, providing strong support for emergency response and rescue. By real-time monitoring and analyzing the changes in gas concentration, it can guide rescue personnel to quickly locate the accident site and take effective measures for rescue and salvage.

Conclusion

• (1) Based on the T-B cell principle, the abnormal gas outbursts adaptive recognition algorithm achieves adaptive recognition of abnormal gas outbursts feature vectors by adapting the detector to adjust and learn vectors through cloning and mutation, enabling the recognition and memory of known or unknown feature vectors of abnormal gas outbursts in dynamic changing environments.

• (2) Based on the analysis of the different characteristics of gas concentration changes under various influencing factors, the DTW algorithm is selected as the abnormal gas outbursts type identification algorithm. According to the principle of biological immune recognition, a abnormal gas outbursts type identification model based on the DTW algorithm is constructed.

• (3) Based on the abnormal gas outbursts anomaly degree recognition algorithm based on the immune algorithm and the abnormal gas outbursts type recognition algorithm based on the DTW algorithm, and combined with the characteristics of the biological immune system, a abnormal gas outbursts risk identification model based on the immune mechanism is constructed, which has the excellent characteristics of adaptability, learning, and memorization.

• (4) Taking the example of a abnormal gas outbursts at the 9111 working face of a certain mine in Huainan, the characteristic vector of abnormal gas outbursts was determined, and an abnormality recognition model based on immune algorithm was input to output the recognition result adaptively; on the basis of recognizing the outburst abnormality, the gas concentration sequence was input to a abnormal gas outbursts type recognition algorithm based on DTW algorithm, and the recognition result was obtained through distance calculation with the pre-collected reference sequence.

• (5) Based on the biological immune system, the risk identification of abnormal gas outbursts in coal mines provides new concepts and ideas, and also lays a certain foundation for the research on risk prevention and control of abnormal gas outbursts. At the same time, by drawing lessons from the relevant mechanisms of the immune system in defending antigens, further studies can be conducted on the construction of a characteristic database and a countermeasure database for abnormal gas outbursts.

• (6) The validity of the model constructed based on the verification of a abnormal gas outburst at the 9111 working face of a mine in Huaibei has certain limitations. It is suggested to further carry out the verification of abnormal gas outbursts risk identification in different regions and work locations.