Stability analysis and predictive modeling of mine pressure manifestation along the air-retained roadway

In traditional shaft mining, coal pillars of a certain width are often left in adjacent faces to control the movement of the overlying rock strata and to isolate the mining airspace, however, a large number of coal pillars are left, resulting in a serious waste of coal resources and an uncoordinated situation of mining and extraction succession in the mines [1–3]. With the reduction of shallow coal seam resources, the depth of coal seam mining is increasing, the mine pressure is more intense, and the width of coal pillar is increasing, which is a serious waste of coal resources [4–5]. As an important technology of coal pillar mining system, the advantage of leaving the roadway along the air is that it can recover the residual coal, edge coal and coal pillar, reduce the amount of roadway excavation, improve the recovery rate of coal resources, and extend the service life of the mine [6–8]. At the same time, along-the-airway technology can also optimize the traditional ventilation method, reduce the accumulation of gas in the working face and the risk of protrusion, etc. [9–10]. Along the air retaining technology is a safe and efficient mining technology, which is of great significance for realizing the scientific and green development of the coal industry.

The increase in the length of the coal seam working face can reduce the workload of the working face back to the mining roadway, reduce the number of times that the comprehensive mining working face is moved, improve the extraction rate of coal resources, maximize the utilization rate of the comprehensive mining equipment, and improve the single production of the working face [11–13]. Although the degree of mechanization and intelligence of Chinese coal mine comprehensive mining equipment is constantly improving, and the level of working face single production and single advancement has been gradually improved, it is still in the primary stage of intelligent construction, and there are many challenges in the construction of intelligent working face. Including insufficient equipment synergy linkage, poor information interoperability and other problems [14]. One of the important issues is the challenge related to the mining pressure monitoring data, which are transferred between the supports with low efficiency, making it difficult to accurately predict the pressure distribution, timing, and strength of the roof plate in the quarry [15–16]. In addition, the limited efficiency of automatic coordination between coal miners and hydraulic supports requires significant manual intervention [17]. Therefore, based on the in-depth analysis of the mining pressure law of the extra-long working face, the intelligent prediction of the incoming pressure on the roof plate is realized by monitoring the mining pressure data and applying intelligent technology, and the scientific difficulties such as the adaptive coordination of the support group are also solved [18–20].

Along the air retaining tunnel technology is a safe and efficient mining technology, which is of great significance for realizing the scientific and green development of the coal industry. Zhang, Z. et al. showed that the stability of retained-groove along-groove (RGSG) is mainly affected by the movement of rock strata on the roof of the working face of the coal seam, and by simulating the dynamic evolution law of deformation of the RGSG perimeter rock and the supporting stresses in the process of coal-seam mining in order to understand the destabilizing characteristics of the RGSG under different roof conditions, and based on which, pre-cracking technology is used to optimize the roof structure to improve the stability of the RGSG [21]. Yang, H. et al. investigated the deformation and stress distribution of the surrounding rock along the hollow stayed roadway under the disturbance of mining in the working face of the adjacent coal seam, and proposed the stabilization control technique combining pressure relief and anchor support, and the experiments showed that the deformation of the roadway top plate and side wall under this technical support was effectively controlled [22]. Xie, S. et al. discussed the comprehensive control technology of perimeter rock under the distance coal seam ER (GSER) along the empty retaining lane, grouting modification and optimization of the roof structure of the GSER roof, using one beam and three columns to strongly support the roof and control layer, using tie rod pre-tensioning and hydraulic pillars to reinforce the side walls, and using bolts to protect the roof of the coal seam [23]. Chen, D. C. et al. examined the application of flexible formwork concrete wall (FFCW) in the stability control of along-void stayed roadway, and investigated the stability mechanism of the surrounding rock of the working face in the process of coal mine back-mining through numerical simulation, and the experiment showed that the deformation control of the roadway surrounding rock was good and there was no obvious pressure manifestation in the working face [24]. Zhang, Z. et al. proposed a method for determining the critical lagging inlet width of GER-LMH supported by the cusp mutation theory based on the results of the characterization of the deformation, stress, and distribution of plastic zones of the surrounding rock of the large-height-along-vacancy-lagging-hole (GER-LMH), which can satisfy the ventilation requirements of the GER-LMH while guaranteeing the stability of the GER-LMH [25].

In the field of international mine engineering, computer technology, including roof mining pressure prediction, is widely used, providing strong technical support for mine engineering and helping to improve work safety and efficiency. Wang, K. et al. combined the gray theory and neural network algorithm to predict the roof pressure of coal seam working face, and the simulation experimental results show that the prediction efficiency and accuracy of the proposed combined prediction model have been significantly improved, which is of great significance for the prevention of roof safety accidents [26]. Tan, T. et al. summarized the mining pressure mechanism of coal seam roof and constructed a genetic algorithm neural network (GA-BP) starting roof pressure prediction model, and experiments showed that the GA-BP model has a larger coefficient of determination, smaller error, and stronger stability than the BP model, which has a broad application prospect in the process of coal mine safety production [27]. ZENG, Q. et al. proposed a combined Prophet+LSTM model fusing the influence of multiple neighboring supports to predict the rock pressure in the working face, and its support rock pressure prediction results have higher stability and accuracy, which provides methodological insight for the prediction of underground pressure in coal mines [28]. Dong, J. et al. established a multivariate linear regression model with the quantitative relationship between the mining pressures of coal seams as the constraints, and utilized the real-time mining pressure monitoring data for training, and the proposed model was able to quickly predict the value of the roof pressure in the next mining area [29]. Zhang, Y. et al. utilized multi-scale context fusion network for roof pressure prediction, and the multi-scale feature extraction method has very high advantages and performance compared with single feature extraction method, and also considering context information makes up for the problems such as insufficient information, which plays a good performance in the multi-step prediction of roof pressure [30].

This paper establishes a simulation model along the open-air stayed tunnel according to the parameter settings, and carries out numerical simulation calculations in FLAC3D software in order to analyze its stability changes. Through a brief introduction of the definition of mine pressure manifestation and its possible disasters, the necessity of predicting the pattern of mine pressure manifestation is emphasized. Distributed fiber-optic sensors are embedded in the simulation model along the open channel to establish a light data monitoring system. The average change of Brillouin frequency shift of the fiber optic reflects the information change of the overburden rock layer in the mining process, and the pressure situation of the overburden layer is measured according to whether the Brillouin frequency shift change curve appears “spikes” or not. The CatBoost algorithm is used to construct a prediction model for the occurrence of ore pressure, in which the sorting and boosting strategy is utilized to add the a priori terms and weighting coefficients, and a relatively independent model is established for different samples, and the base learner is continuously trained according to the gradient values to realize unbiased gradient estimation. In order to further improve the prediction accuracy of the model for mine pressure manifestation, this paper introduces the Bayesian algorithm to iteratively optimize the prediction model. Through the simulation numerical calculations, this paper finally analyzes the predicted results for the stability of along-void stay channel and ore pressure manifestation.

2

Stability analysis model for along-air stay lanes

The coal seams mined in the working face of a town town bottom mine are No.5 and No.7 seams, which are stable, with the thickness of 2.1~3.6m, the average thickness of 2.9m, and the structure of 1.8(0.3)0.6. The working face as a whole is in the monoclinic structure, which is roughly tilted from east to west, with the maximum angle of inclination of the seams at 10°, the minimum angle of inclination of the seams at 0.8°, and the average angle of inclination of the seams at 4.3°.

The comprehensive release along the air to stay along the roadway is the use of flexible mold concrete on the original roadway for artificial settings and the formation of a special back to mining roadway [31], due to stay along the roadway of the overlying rock, low-strength coal pillars and high-strength flexible concrete strength is very different, so that the deformation of the surrounding rock analysis is more complex, and its deformation characteristics can not be analyzed in the form of the traditional roadway analysis. Along the air stayed the general experience of two mining and the first mining impact, the overburden rock strong movement, making the roadway perimeter rock damage is serious, the roadway is extremely difficult to stabilize. In order to analyze the movement of the surrounding rock along the open channel, FLAC3D numerical simulation software is used to simulate the movement of the surrounding rock.

First of all, the simulation model is established, in order to simplify the simulation process, the flexible mold wall is set as a concrete wall, according to the actual geological conditions, the model size length, width and height of 180, 130 and 38m respectively, the thickness of the overlying rock in the model is 12m, the width of the roadside support along the empty stay lane is set at 1.3m, and the vertical stress of the basic top is set at 10.38MPa. The boundary conditions of the model are set as follows. The left boundary of the model is fixed to move in the X direction, the three-way deformation of the bottom of the model is fixed, and the mechanical parameters of each rock layer of the model are set according to the actual geological conditions, and the mechanical parameters of the rock layer are shown in Table 1. In the table, B, S, V, I, C, T are bulk weight, shear modulus, bulk modulus, internal friction angle, cohesion and tensile strength, respectively.

Table 1.

Mechanical parameters of rock strata

Lithology	B/(kg·m^-3)	S/GPa	V/GPa	I/°	C/MPa	T/MPa
Fine sand	2140	12.31	16.28	32	7.86	4.68
Sandy shale	2380	1.72	2.62	38	4.73	3.82
Coal	2450	0.63	1.17	34	2.19	1.15
Silt sandstone	2320	9.84	14.96	33	6.24	3.69
Mesosandstone	2100	8.92	10.58	31	3.17	2.85
Mudstone	2440	1.04	2.41	35	2.21	1.06

3

Prediction of mineral pressure manifestation

The coal mining process destroys the original stress equilibrium state between the rock layers, and under the influence of gravity, the rock body will generate a new stress field, which is usually called mine pressure manifestation [32]. Intense mine pressure can induce geological disasters such as roadway deformation and instability, gas protrusion, impact ground pressure, etc., which bring huge losses for coal mine production. Therefore, in order to reduce the occurrence of geological disasters, mastering the law of mine pressure manifestation and effectively predicting and controlling the mine pressure is one of the important means for the safe production of coal mines. Therefore, this paper will use CatBoost model optimized based on Bayesian algorithm to predict the appearance of mine pressure.

3.1

Fiber optic data monitoring system construction

Distributed fiber optic sensors are buried in the model constructed in the previous section to establish a fiber optic data monitoring system as shown in Figure 1. The system consists of fiber optic sensors, NBX-6055 stress analyzer and computer. In the middle of the model, three vertical optical fibers Fv1, Fv2, and Fv3 are equally spaced, and the sampling interval of the optical fiber data is 20 mm, and the effective length is 1750 mm. When simulating the digging of the working face, the optical fibers are subjected to the deformation force of the rock layer, and then frequency shift occurs, and then the optical signals are converted into electrical signals by the NBX-6055 Stress Analyzer and transmitted to the computer, and then finally the computer software is used to realize the analysis and storage of the optical fiber data, and the optical fiber data are collected. The fiber optic data are parsed and stored, and the collected data are stored in .bat format. In order to facilitate the processing, the data is migrated to excel table, the data structure is in the form of 175×50 matrix, the rows represent the data of each position of the whole optical fiber, and the columns represent the advancing distance of the working face.

3.2

Predictive modeling of mineral pressure manifestation

3.2.1

Characterization of mineral pressure by fiber-optic Brillouin frequency-shift averaging variability

In the process of using distributed fiber optics to monitor the mining pressure, the information change of the overlying rock layer during the mining process can be reflected by the change of the Brillouin frequency shift [33]. The Brillouin frequency shift basically does not change significantly before the excavation of the coal seam, while in the excavation process, the Brillouin frequency shift will produce more obvious changes when the overlying rock layer is deformed and damaged. The degree of change of Brillouin frequency shift is positively correlated with the degree of deformation and damage of the overlying rock layer, and based on this law, the expression for the average degree of change of fiber-optic Brillouin frequency shift is deduced: 1 $D_{l} = \frac{1}{n} (\sum_{i = 1}^{n} | B_{i} | - \sum_{i = 1}^{n} | B_{i - 1} |)$

Where: D_i is the average variation of the Brillouin frequency shift of the fiber at the excavation distance l. n is the total number of sampling points on the fiber. B_i is the Brillouin frequency shift of the i(i = 1,2,…,n)th sampling point on the fiber at each excavation. The degree of deformation of the overlying rock layer at a certain time period can be expressed by the difference between the Brillouin frequency shifts of two adjacent times before and after that time period. The Brillouin shift change curve will appear “sharp peaks” as the excavation progresses, which can be used as a judgment indicator to measure whether the overlying rock layer is under pressure or not.

3.2.2

CatBoost Prediction

The fiber Brillouin frequency shift variability dataset is preprocessed by using Takens phase space reconstruction theory, one-dimensional projection of the dataset, determining the delay time τ by the mutual information method, and determining the embedding dimension m by the method of minimizing the prediction error, and converting the dataset into a time series that varies with the excavation distance, X ={X₁,X₂,…,X_n–1}, where X_i = {x_i,x_i+1,…,x_i+(m+1)τ}, x_i is the Brillouin shift variability of the i(i =1,2,…,n-1)th sampling point on the fiber. The samples after phase space reconstruction are: 2 $x_{i + 1 + (m - 1) τ} = f (x_{i}, x_{i + 1}, \dots, x_{i + (m - 1) τ})$

The mapping function f(·) is obtained through learning so that Y of the test set is mapped to Y, Y = {Y₁,Y₂,…Y_n–1} of the training set through the mapping function, where the error between Y_i = x_{i+1+(m–1)_T}, Y and the true value should be as small as possible.

CatBoost, as an upgraded GBDT algorithm [34], makes the following improvements over the traditional GBDT algorithm. 1)

Since the GBDT algorithm uses the average value of labels to measure whether a node is split or not, this measurement condition leads to information missing to some extent, which leads to gradient bias and prediction bias problems. To solve this problem, the CatBoost algorithm utilizes a rank-boosting strategy, adds a priori terms and weight coefficients, establishes relatively independent models for different samples, and continuously trains the base learner based on the gradient values to achieve unbiased gradient estimation: 3 ${\hat{ξ}}_{k}^{γ} = \frac{\sum_{i = 1}^{n} I_{{ξ_{i}^{γ} = ξ_{i}^{γ}}} y_{i} + a p}{n \sum_{i = 1}^{n} I_{{ξ_{i}^{γ} = ξ_{i}^{γ}}} + a}$ 4 $I_{{ξ_{i}^{r} = ξ_{i}^{r}}} = {\begin{matrix} 1 & ξ_{i}^{γ} = ξ_{k}^{γ} \\ 0 & other \end{matrix}$

Where: $ξ_{k}^{γ}$ is the γ rd category feature of the k nd training sample. ${\hat{ξ}}_{k}^{γ}$ is the unbiased gradient estimate of $ξ_{k}^{γ}$ . I is the indicator function. $ξ_{i}^{γ}$ is the γ th category feature of the i th sample point data for each excavation. y_i is the label of the i th sampling point data. a is the weight coefficient. p is the prior term.

2)

The CatBoost algorithm chooses a forgetful decision tree, which ensures that the split features of each layer are consistent in the iterative process, and the left and right sub-trees are completely symmetric and balanced, which reduces the complexity of the model and thus improves the prediction accuracy, the training speed, and the memory utilization.

3.2.3

Parameters of the Bayesian optimization algorithm

In order to further improve the accuracy of CatBoost algorithm in mine pressure prediction, the parameters of CatBoost algorithm are optimized during the training process, and this paper optimizes the parameters by Bayesian algorithm, which is mainly divided into the following 2 steps. 1)

Assuming that the training set is M, the set of unknown objective functions is {g(η₁),g(η₂),…,g(η_n–1)}, η_i is the samples in the training set M, and the hyperparameters are θ. When there is observation noise ε and assuming that the noise meets the Gaussian distribution of independent homogeneous distributions P(ε) = (0,σ²) (σ² is the predicted variance of the a priori model), the marginal likelihood distribution can be obtained: 5 $P (η_{i} | M, θ) = \int_{θ} P (η_{i} | g (θ)) P (θ_{best} | M) d θ$ where θ_best is the optimal solution for the observations. θ_best is obtained by maximizing the marginal likelihood distribution by great likelihood estimation.

2)

Choose a collection function and construct a utility function from the posterior model to determine the next collection point. In this paper, we use the commonly used expectation boosting function to find the maximum value of the expectation increment in the current optimal case. The acquisition function is: 6 $α (θ | μ, σ) = E [\max (0, g (θ) - g (θ_{best}))]$

Where: μ is the predicted mean of the a priori model. E(·) is the expectation at the current optimum.

4

Results and analysis

4.1

Stability analysis along the air stay lane

4.1.1

Stability Analysis of Coal Gangs along the Air Stay Lane

According to the record of measuring point of the roadway, taking the distance of the measuring point from the coal wall of the working face as the horizontal coordinate, the horizontal stress and displacement of the coal gangs along the hollow stay roadway are plotted as shown in Fig. 2 and Fig. 3, respectively, with the advancement of the working face. It can be seen that the curves to the left and right of the horizontal coordinate 0 point in the figure are the stress-strain law of the roadway 30m in front of the working face and 70m behind the working face respectively. Generally speaking, the horizontal stress of the coal gang is mainly tensile stress, which is cyclic in the whole back-mining roadway, and the pressure in a certain range of the roadway in front of and behind the coal-mining work is frequent and intense, and tends to moderate with the working face away from the stress. Specifically, the horizontal stress of the coal gang suddenly increased to 2.364MPa and 4.702MPa at 5.105m and 17.080m of the roadway in front of the coal mining workings, which is mainly due to the stress change caused by the cyclic activity of the overlying rock layer on the workings, and the cycle of the coal gang comes to the pressure with a step of 13m. The horizontal stress of the coal gangs within 3~20m from the front of the mining face continues to be 1.0MPa, which indicates that the coal gangs in this section of the roadway are the key supporting parts when over-supporting. After the coal mining face 0 ~ 50m section of the coal gangs along the empty stay lane more frequent and intense pressure, especially in the stay lane from the mining face 0.8m at the coal gangs pressure increased sharply to 6.31MPa, and this position is in the concrete wall casting construction section along the empty stay lane. The average stress of 1.0~1.5m section is maintained at about 3.50MPa, and then the stress tends to slow down, mainly oscillating between 0.0~2.0MPa, which indicates that compared with the original roadway, the coal gangs along the open-air stayed roadway are more intense and the step distance of the pressures is shorter (8m on average), which is mainly due to the result of the joint influence of the coal mining activities and the stayed roadway support construction, and indicates that the stability of roadway gangs within the 50m range (especially at the 10m after mining) of the stayed roadway after mining should be emphasized. This indicates that we should pay attention to the stability of the roadway gangs within 50m after mining (especially at 10m after mining). The displacement of the gangs along the open-air stayed alley continued to increase with the distance from the mining face, and the displacement of the gangs increased from 0.994mm to 2.869mm in the 0~15m section, and then the increase of the displacement slowed down, and the displacement of the gangs increased to a maximum of 5.936mm in the whole monitoring range, which indicated that the use of the concrete filler wall support along the open-air stayed alley was favorable to the control of the gangs' displacement.

4.1.2

Analysis of the stability of the roof slab along the air-retaining tunnel

The changes of vertical stress and roof subsidence of the roof plate along the hollow stay are shown in Fig. 4 and Fig. 5, respectively. The distance from the working face is the horizontal coordinate, and the stress and displacement are the vertical coordinates. From the figure, it can be seen that the roof plate of the whole back-mining tunnel is mainly subjected to vertical compressive stress. The roadway roof at 9.516m in front of the working face was first subjected to a tensile stress of 6.256MPa, after which the tensile stress was immediately changed to a compressive stress of 15.486MPa, which was caused by the sudden coming pressure on the roadway roof due to the subsidence, fracture and rebound of the rock strata caused by the coal mining work. The roadway roof at the measuring point 0 was pressurized the most, reaching 23.925MPa, which was exactly at the end of the working face, and it was the position where the key support was necessary. 0~32 m range of the roof along the empty stayed roadway came to the pressure more intense, the pressure cycle was obvious, the average pressure was around 2.529MPa, and the roof of the roadway outside of 32m tended to be moderated, and the pressure was oscillating between 0.995~3.428MPa. The pressure oscillates between 0.995~3.428MPa. This indicates that as the roof slab in the hollow zone sinks, collapses and compacts, the roof slab activity along the hollow stay tends to moderate. The subsidence of the roof of the roadway in front of the working is very small, but the subsidence of the roof of the roadway after mining is obvious, especially the subsidence speed of the roof of the roadway outside the mining face of 40m in the roadway increases, and the maximum roof subsidence reaches 77.605mm within the monitoring range, which may be due to the joint action of the roof collapse of the goaf and the concrete wall support, the roof of the roadway is cut or its cracks and joints can be developed in large quantities, which further destroys the integrity of the roof of the roadway and reduces the strength of the roof, so the roof of the roadway is pressed and relaxed. It can still cause a large roof subsidence, indicating that in addition to the original roadway support, anchor cables, beam-column supports, and even grouting should be added to strengthen the stability of the roof of the roadway.

4.2

Prediction of the pattern of mineral pressure manifestation

4.2.1

Model evaluation

In order to make the prediction effect more widely applicable, the model performance is evaluated using MSE, root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination R². Smaller values of MSE, RMSE, MAE and larger values of R² indicate higher accuracy of the model prediction.The formula for MSE, RMSE, MAE, R² are calculated as follows, respectively: 7 $E_{MS} = \frac{1}{m} \sum_{i = 1}^{m} {(Y_{i} - {\bar{Y}}_{i})}^{2}$ 8 $E_{RMS} = \sqrt{\frac{1}{m} \sum_{i = 1}^{m} {(Y_{i} - {\tilde{Y}}_{i})}^{2}}$ 9 $E_{MA} = \frac{1}{m} \sum_{i = 1}^{m} | Y_{i} - {\tilde{Y}}_{i} |$ 10 $R^{2} = 1 - \frac{\sum_{i = 1}^{m} {(Y_{i} - {\tilde{Y}}_{i})}^{2}}{\sum_{i = 1}^{m} {(Y_{i} - \bar{Y})}^{2}}$

Where: m is the sample size. Y_i is the i rd sample true value. ${\tilde{Y}}_{i}$ is the i th sample predicted value. $\bar{Y}$ is the average of the true values.

4.2.2

Data set partitioning

The training set division has a great impact on the prediction performance of the model. For the processed data, the training set with different occupancy ratios is input into the CatBoost (B-CatBoost) model optimized based on Bayesian algorithm for training, and the results of the training volume are shown in Table 2. According to the results in the table, the various errors of the model test results are larger when the training set percentage is %, and the various errors are minimized when the percentage is 90%. Therefore, the processed data is divided into training set and test set by 9:1.

Table 2.

B-CatBoost model training results

Proportion/%	MSE	RMSE	MAE
50	48.36	8.71	6.82
70	10.29	5.47	3.25
90	2.84	1.48	1.21

4.2.3

Predicted effects and performance

1)

Comparison of model prediction effect

In order to test the performance of the B-CatBoost model, based on the simulation model coal wall gang measurement point data, the training set and test set were divided according to the same proportion, and the traditional long short-term memory network (LSTM) model and CatBoost model were used to make predictions, and the prediction results were compared with those of the B-CatBoost model, respectively. The prediction results of each model are shown in Fig. 6, and (a)(b)(c) denote the prediction results of LSTM, CatBoost, and B-CatBoost models, respectively. From Fig. 6(a), it can be seen that the prediction trend of the LSTM model is consistent with the trend of the test set data, but the error between the predicted and actual values is large. From Fig. 6(b), it can be seen that the CatBoost model predicts better for the data with mine pressure lower than 30 MPa, but worse for the data with pressure peak or sudden pressure change. From Fig. 6(c), it can be seen that the B-CatBoost model has the best fit between the predicted and actual values and the highest model accuracy.

2)

Model performance comparison

The error comparison of B-CatBoost model with CatBoost and LSTM model is shown in Table 3. From the table, it can be seen that the LSTM model has higher and least accurate error metrics than CatBoost and B-CatBoost models. Compared with the unoptimized CatBoost model, the B-CatBoost model reduces the MSE by 92.96%, the RMSE by 78.22%,the MAE by 84.54%, and the R² improves by 67.80%, which indicates that the B-CatBoost model is better fitted in the training and testing process.

Table 3.

Error comparison of each model

Model	MSE	RMSE	MAE	R²
LSTM	11.43	6.87	5.84	0.32
CatBoost	5.97	3.26	3.17	0.59
B-CatBoost	0.42	0.71	0.49	0.99

In order to verify the applicability of the B-CatBoost model in the prediction of mine pressure manifestation law, the coal pillar gang data of No. 1, No. 2 and No. 3 stations in the simulation model are used as the data source of the prediction model for verification, and the results are shown in Fig. 7, with the prediction results of No. 1, No. 2 and No. 3 stations shown in (a), (b) and (c) respectively. It can be seen that for the mine pressure data of the coal pillar gangs in the working face of a Zhenchengdi mine selected in this paper, the prediction with the B-CatBoost model has a high prediction accuracy, and the actual value is consistent with the predicted value.

The prediction errors of the B-CatBoost model for the coal pillar gang data are shown in Table 4. It can be seen that the fit of the model to the mine pressure data of No.3 station reaches 0.998, and the MSE, RMSE and MAE are all reduced to less than 0.15, and the fit to the coal pillar gang data of No.1 and No.2 stations also reaches more than 0.985, which indicates that the B-CatBoost model has a good nonlinear fitting ability and generalization ability.

Table 4.

Prediction of each test station

Test station	MSE	RMSE	MAE	R²
Teat station 1	0.17	0.28	0.32	0.986
Teat station 2	0.14	0.21	0.11	0.987
Teat station 3	0.11	0.12	0.09	0.998

5

Conclusion

This paper establishes a simulation model along the open channel through numerical simulation software and analyzes the stability of the open channel according to numerical calculations. Based on the Bayesian optimization algorithm, the B-CatBoost model is constructed to predict the manifestation of mine pressure. 1)

The horizontal stress of the coal gang is mainly tensile stress, and it changes periodically in the whole back-mining roadway. The horizontal stresses of the coal gangs at 5.105m and 17.080m from the mining face are 2.364MPa and 4.072MPa respectively, with a drastic increase. After 0~50m section of coal mining face, the pressure of coal gangs along the open road is more frequent and intense, and the pressure of coal gangs in the open road increases sharply to 6.31MPa at 0.8m from the mining face; meanwhile, the displacement of coal gangs along the open road increases with the increase of the distance from the mining face, and the displacement of coal gangs increases from 0.994mm to 2.869mm within 0-15m distance with the growth rate of 188.63%. It has a greater impact on the stability of the open-air stay lane. The maximum pressure on the roof plate along the empty stay can be up to 23.9MPa, and the maximum sinking amount is 77.065mm, which also needs to take support or reinforcement measures to ensure the stability of the alley along the empty stay.

2)

In the prediction of mine pressure manifestation, the model of this paper has the best fitting effect between the predicted and actual values, and the model has the highest prediction accuracy. In each performance evaluation index, the MSE, RMSE, and MAE of this paper's model are reduced by 96.33%, 89.67%, 91.61%, 92.96%, 78.22%, and 84.54% compared with the LSTM model and the unoptimized CatBoost model, respectively. The fitting coefficient R² is improved by 209.38% and 67.80% compared to the other two models, respectively. It verifies the good performance of the model of this paper in the prediction of mine pressure manifestation. The errors of the model in predicting the mine pressure of the coal pillar gangs at the three stations are all below 0.5, and the fitting coefficients R² are all above 0.95, which further indicates that the model in this paper is able to excellently fulfill the task of predicting the mine pressure manifestation.

Idioma:: Inglés

Calendario de la edición:: 1 veces al año
Temas de la revista:: Ciencias de la vida, Ciencias de la vida, otros, Matemáticas, Matemáticas aplicadas, Matemáticas generales, Física, Física, otros

RSS Feed de revista

Stability analysis and predictive modeling of mine pressure manifestation along the air-retained roadway

Rui Liu

Publicado en línea: 19 mar 2025

Recibido: 09 oct 2024

Aceptado: 29 ene 2025

DOI: https://doi.org/10.2478/amns-2025-0514

Palabras claveAlong the open channel, Mine pressure manifestation, CatBoost model, Bayesian algorithm

© 2025 Rui Liu, published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Palabras clave
Along the open channel, Mine pressure manifestation, CatBoost model, Bayesian algorithm