Fuzzy rule based classification method of surrounding rock stability of coal roadway using artificial intelligence algorithm

Abstract

As the stability of surrounding rock of coal roadway is affected by many factors, which makes the classification result hard to be consistent with the field practice. To solve the above problems, this paper proposes a method for the classification of stability of rock which is present in roadway of coal using the artificial intelligence algorithm. In this paper, the influencing factors of stability of rock which is present in roadway are analyzed, and seven influential factors are selected as classification indexes. To solve the problem of slow convergence speed and easy to fall into the local minimum of the back propagation artificial neural network (BP-ANN), an improved BP-ANN algorithm based on additional momentum and Levenberg-Marquardt optimization is proposed based on the analysis of the existing improved methods, which improves the convergence speed and avoids the local minimum effectively. Based on the learning model available, classification system based on fuzzy rule have been implemented and yielded better behavior in the situation of uncertain data sets. Finally, the stability classification model of surrounding rocks of coal roadway using BP-ANN was established in MATLAB environment, and the model was applied to 13 data samples of coal roadway for testing, with the identification rate of 92.3%. The experimental results verify that the method proposed based on fuzzy rule classification system in this paper has a high accuracy of type identification and is applicable to the stability classification of surrounding rock in the coal roadway.

Keywords

Coal roadway surrounding rock artificial intelligence BP-ANN fuzzy rule classification system

1 Introduction

The classification of stability of rock formed in coal runway is the basis of reasonable roadway support design and the important premise of predicting the stability in rock covering the roadway of coal in the exploitation and removal process [1]. Therefore, the scientific and accurate classification of surrounding stability of rock which is present in roadway will directly affect the support form and method, thus affecting the support effect of the coal roadway [2]. So the classification of stability of rock which is present in roadway must be solved first. The accurate, scientific, and convenient classification of surrounding rock stability is always an important problem for mining experts and scholars [3, 4].

In recent years, because of the complex geological conditions of surrounding rock and the numerous factors affecting its stability, researchers have widely used mathematical statistics, fuzzy mathematics and other modern mathematical methods in the classification of stability [5 –9]. The classification of surrounding rock stability changes from qualitative to quantitative and various classification methods are put forward to evaluate surrounding rock stability comprehensively and objectively. However, as the surrounding rock is a complex medium, there is a complex nonlinear relationship between its stability and influencing factors, so the general method cannot truly reflect the nonlinear relationship. Although the present research has made some achievements, it has not been widely applied and still needs further improvement [10, 11].

Due to the complexity of geological conditions and limited geological prospecting methods, it is not possible to obtain all the data to determine the grade of surrounding rocks accurately. Therefore, the traditional classification of surrounding rocks can only be determined by experience and qualitative methods. With the application of various new methods and technologies in the field of mining, a variety of methods have emerged for the stability classification of coal roadway surrounding rocks, including fuzzy theory, grey theory, expert system, abrupt progression method, artificial neural network, and other intelligent prediction methods [12 –14]. These methods effectively improve the accuracy of classification, but they all have some shortcomings. In fuzzy theory, membership degree and weight are difficult to determine. The accuracy and simplicity of the grey theory are not enough. The expert knowledge used in expert systems is often trivial, imprecise, and uncertain; the method of abrupt progression is more complicated in the classification of surrounding rock stability. The randomness of surrounding rock instability cannot be considered by the artificial neural network. Because of the influence of random factors, probabilistic neural network and immune evolutionary algorithm are combined in [15] to achieve accurate classification of the stability of slope surrounding rock. In [16, 17], the probabilistic neural network method was used to detect the damage of the bridge panel of the suspension bridge, and the test results showed that the probabilistic neural network method was feasible and effective.

The artificial neural network, especially BP-ANN, has a good modeling ability, which can truly reflect the nonlinear relationship between the problem and its influencing factors without the need for the specific functional form. Moreover, the neural network has the advantages of parallel information processing, distributed storage, self-organization, strong nonlinear approximation, and so on. Therefore, BP-ANN has a great advantage in dealing with the nonlinear classification of stability found in rocks of the coal roadway. However, the traditional BP-ANN has some disadvantages, such as local minimum, the slow convergence speed of learning algorithm, and difficulty in determining network structure. Therefore, it is necessary to optimize and improve the traditional BP-ANN [18, 19].

Because of the above problems, this paper proposes a classification technique of nearby rock stability of coal roadway on improved BP-ANN. The contents of this paper include the following aspects: firstly, this paper analyzes the influencing factors of surrounding rock stability of coal roadway, and selects seven factors that have a great influence on surrounding rock stability as the classification indicators; then, the principle of BP-ANN is studied. Aiming at the shortcomings of BP-ANN, like lower speed convergence and easy to handle a minimum local, an improved BP-ANN method based on additional momentum and optimization is proposed, which improves the convergence speed of the network and avoids local minimum effectively. Finally, the stability classification model of surrounding rocks of coal roadway using BP-ANN was established in MATLAB environment [20], and the model was applied to 13 data samples of coal roadway for testing, with the identification rate of 92.3%. The experimental results verify that the method proposed in this paper has a high accuracy of type identification and is applicable for classification of nearby stability of rock which is found in roadway. Fuzzy Rule-Based Classification System (FRBCSs) are helpful and well established application in machine learning plat form. They offer significant trade off amongst the empirical exactness of techniques belonging to traditional systems and the achieved interpretability utilizing the labels of linguistic which contains sematic closer to natural language, earlier works has provided the FRBCSs having better behavioral knowledge with uncertain data sets by using constant preprocessing techniques.

The hybridization amongst genetic algorithm and fuzzy logic modelling to Genetic Fuzzy Models (GFMs), which is vitally utilized methods when various computational intelligence methodologies are associated. A GFM is normally a fuzzy method formed by learning methodologies on the basis on evolutionary models. In various evolutionary algorithms, the improvement of Genetic Programming (GP) of modernized genetic algorithms that involve tree shaped formulation utilizing chromosome which has the variable length. GP handles the utilization of FRBCSs to form rule based learning algorithm based on fuzzy. Moreover, the certain lack of FRBCSs is the concepts inflexibility of using linguistic components because it implements tough restrictions on structure of fuzzy rule, which lacks in terms of accuracy when handling complex issues, like huge dimensional issues, the attendance of noise or classes which is overlapped. Various possibilities to improvise the model of linguistic fuzzy have been incorporated.

2 Related work

Liu and Chen [21] mentioned a unique model for classification of rock mass on finding the stability of rock slope on the basis of analyzing through hierarchy process, linear discriminant and fuzzy logic model. Liu et al. [22] developed the fuzzy probabilistic process for estimating the depth stability while slope cutting in rock mass. They utilized fuzzy models to assist traditional approaches for stability achievements. Wu et al. [23] mentioned fuzzy set model analysis to improvise indeterminate issues in estimating the discrepancy coefficient for finding stability of slopes cutting in rocks. Fang and Shen [24] have implemented decision making group using multi attribute on basis of fuzzy. Mikaeil et al. [25] utilized fuzzy and multi criteria decision forming methods for calculating of saw ability in rocks in mining fields.

Jato-Espino et al. [26] mentioned a survey article based on decision making using multi criteria application in civil works. Su et al. [27] demonstrated the fuzzy theory of optimal recognition for stability in slope estimation in hydropower plants. Rafiee et al. [28] developed the fuzzy model for rock modelling improvement, which allows the deliberation of doubt on basis on classical expert methodology which were utilized to estimate cavability in sediment of rocks. Khakestar et al. [29] moderated by applying decision making models on geotechnical components. Mirhossein et al. [30] estimated the fuzzy possibility decision making to praise the key set model in rock analysis for slope stability in rocks.

3 Influencing factors of surrounding rock stability of coal roadway

Table 1 shows the list of influencing components of nearby stability of rock in roadway of coal. These factors have different effects on the stability of surrounding rock, which can be divided into the engineering geological factors and mining technology factors. Engineering geological factors include the strength and structural characteristics of surrounding rocks, the buried depth of roadway, the characteristics of coal seams, the greatness and way of in-situ stress, groundwater and ore pressure, etc. The mining technical factors include roadway width, height, and width of the chain pillar.

Table 1
The factors influencing the stability of coal road surrounding rock

No. Type Factor No. Type Factor

1 intensity of the uniaxial compressive strength of direct roof 12 geologic parameters of coal layer thickness

2 surrounding rock Strength of old roof 13 dip direction

3 intensity of coal layer 14 buried depth

4 strength of direct bottom 15 characteristics of Top and bottom thickness of direct top

5 coal-mining method 16 thickness of old roof

6 ground stress size 17 thickness of direct bottom

7 angle with roadway 18 characteristics of ground pressure direct top first caving step

8 width of roadway pillar 19 old roof second caving step

9 characteristics height 20 old roof period press step

10 of roadway width 21 underground water

11 buried depth

No.	Type	Factor	No.	Type	Factor
1	intensity of the	uniaxial compressive strength of direct roof	12	geologic parameters of coal layer	thickness
2	surrounding rock	Strength of old roof	13		dip direction
3		intensity of coal layer	14		buried depth
4		strength of direct bottom	15	characteristics of Top and bottom	thickness of direct top
5	coal-mining method		16		thickness of old roof
6	ground stress	size	17		thickness of direct bottom
7		angle with roadway	18	characteristics of ground pressure	direct top first caving step
8	width of roadway pillar		19		old roof second caving step
9	characteristics	height	20		old roof period press step
10	of roadway	width	21	underground water
11		buried depth

Based on the principle of classification index and the comprehensive estimation of components on stability of rock present in roadways, this article selects the following seven indicators as indicators of stability classification of surrounding rock of roadway, including roof strength, coal seam strength, floor strength, buried depth of roadway, the ratio of the mining height, roadway pillar width, direct top first caving step.

4 Improved BP-ANN algorithm

4.1 Principle of BP-ANN

The BP-ANN is included with an input seam, one or more hidden layers, and output seam. Each layer is composed of many simple neurons with parallel operations. The neurons between the network layer and layers adopt full interconnection mode, and there does not includes mutual association amongst the similar layer neurons. Although the structure of a single neuron is simple and its functions are limited, the network system composed of a large number of neurons can achieve extremely powerful functions. Although the structure of a single neuron is simple and its functions are limited, the network system composed of a large number of neurons can achieve extremely powerful functions. The structure of BP-ANN is shown in Fig. 1.

Fig. 1

Structure of BP-ANN.

The learning and training of BP-ANN algorithm is divided into three stages: forward propagation stage, error backpropagation stage, and weight update stage. The specific process of the three stages is as follows.

(1) Onward spread stage

The output of the input layer is: $o_{i}^{(1)} = x (i) (i = 1, 2)$ (1)

It can be seen from the Equation (1) that the input value of ANN is the output value of the input layer of the ANN.

The initial and final seam are described as follows: ${net}_{j}^{(2)} = \sum_{j = 1}^{3} w_{ij}^{(2)} o_{i}^{(1)}$ (2)

$o_{i}^{(2)} (k) = f ({net}_{j}^{(2)} (k))$ (3)

Here, $w_{ij}^{(2)}$ is the weight coefficient of the hidden layer. In the upper corner, (1) represents the input layer; (2) represents the hidden layer; (3) represents the output layer. f() refers to the activation function of the hidden seam neuron, f() depicts sigmoid purpose with positive and negative symmetry.

The input and output of the output layer are as follows: ${net}_{l}^{(3)} = \sum_{l = 1}^{5} w_{lj}^{(3)} o_{j}^{(2)}$ (4) $o_{l}^{(3)} (k) = g ({net}_{j}^{(3)} (k))$ (5)

Here, $w_{lj}^{(3)}$ is the weight coefficient of the output layer, g() is the activation function of the output layer, and g() is the non-negative sigmoid function.

(2) Error backpropagation stage

After the forward propagation phase, the calculated mean square value of the error function is: $J = {\frac{1}{2}}^{2}$ (6)

The calculated error mean square value is corrected by the negative gradient to the weight coefficient of the hidden layer and the output layer so that the increment of the weight coefficient of each layer is: $w_{lj}^{(3)} (k + 1) = - η \frac{\partial J}{\partial w_{lj}^{(3)}} + α Δ w_{lj}^{(3)} (k)$ (7)

Here, η for the rate of the learning, α is the coefficient of inertia.

(3) Weight update stage

After the error backpropagation, the weight coefficients of the hidden layer and the output layer will be adjusted according to the negative gradient of the objective function. $Δ w_{lj}^{(3)} (k + 1) = - η \frac{\partial J}{\partial w_{lj}^{(3)}} = - η \frac{\partial J}{\partial {net}_{l}^{(3)}} o_{j}^{(2)}$ (8)

Definition: $δ_{l}^{3} = - η \frac{\partial J}{\partial {net}_{l}^{(3)}} = e (k + 1) \times o_{l}^{(3)} \times (1 - o_{l}^{(3)})$ (9)

Therefore, the weight coefficient correction of the output layer is: $Δ w_{lj}^{(3)} (k + 1) = - η \times e (k + 1) \times o_{l}^{(3)} \times (1 - o_{l}^{(3)}) \times o_{j}^{(2)}$ (10)

Here, η is the rate of the learning, $o_{j}^{(2)}$ is the yield of the hidden seam, e(k + 1) depicts initial deviation value of the controller at the sampling time of k + 1, and $o_{j}^{(3)}$ is the output seam.

The modified increment formula for the weight coefficient of the output layer is: $w_{lj}^{(3)} (k + 1) = w_{lj}^{(3)} (k) + Δ w_{lj}^{(3)} (k + 1)$ (11)

Similarly, according to the gradient method, it can be known that the adjustment parameter coefficient of weights of the hidden seam of BP-ANN is: $Δ w_{ij}^{(2)} (k + 1) = - η \frac{\partial J}{\partial {net}_{j}^{(2)}} o_{j}^{(1)}$ (12)

Definition: $δ_{j}^{3} = - \frac{\partial J}{\partial {net}_{j}^{(2)}} = - \frac{\partial J}{\partial o_{j}^{(2)}} \times o_{j}^{(2)} \times (1 - o_{j}^{(2)})$ (13)

The modulation parameters of the hidden layer weight coefficient are: $Δ w_{ij}^{(2)} (k + 1) = η \times \sum_{l = 1}^{3} (δ_{l}^{3}) \times w_{lj}^{(3)} \times o_{j}^{(2)} \times (1 - o_{j}^{(2)})$ (14)

The modified increment of the weight coefficient of the output layer is: $w_{ij}^{(2)} (k + 1) = w_{ij}^{(2)} (k) + Δ w_{ij}^{(2)} (k + 1)$ (15)

4.2 Improved algorithm

For finding the issues of lower speed convergence and informal to the local lower of BP-ANN, an improved BP-ANN algorithm based on additional momentum and Levenberg-Marquardt optimization is proposed.

(1) Method of additional momentum

The addition of momentum process is to involve the influence of final weight modification using the factor of momentum, that is, to add momentum term when adjusting the weight. $Δ w_{ij} (k + 1) = η \frac{\partial E}{\partial w_{ij}} + α Δ w_{ij} (k)$ (16)

Here, α is the momentum coefficient, 0 < α < 1, Δw_ij(k + 1) is the value of the weight correction at this time, Δw_ij(k) is the weight of the last time, E is the mean sum of errors.

It can be seen from Equation (16) that the change of weight is related to the direction of updating. If the updating direction of the weight is the same as the previous step, then the weighted sum value increases to make Δw_ij(k + 1) increase, thus speeding up the adjustment speed during the stable adjustment. If the updating direction of the weight is different from the previous step, then the weighted sum value decreases to make Δw_ij(k + 1) decreases, which plays a stabilizing role. This method not only considers the effect of the error on the gradient but also reduces the sensitivity of the network in local error regulation, so as to prevent the network from falling into local minima and reduce the oscillation in the learning process.

(2) Optimization algorithm of Levenberg-Marquardt

The Levenberg-Marquardt algorithm is a combination of the Gauss-Newton method and gradient descent method, which has both the locality and rapidity of the former and the global and convergence of the latter, greatly improving the generalization ability and convergence speed of the network. The weight updating formula is: $w (k + 1) = w (k) - (J_{k}^{T} J_{k} + μ I)^{- 1} J_{k}^{T} e_{k}$ (17)

Here, e_k is the error vectors for computing and back-propagation network training; J_k refers to the Jacobian-Matrix of the differential of network learning error concerning weights in the kth iteration, which contains weights and thresholds and is the first derivative of network training error; w(k) is the network weight in the k^th iteration; μ is a scalar; I is the given sample matrix.

When μ goes down to 0, it becomes Newton’s method of approximating the Hessian matrix; When μ increases to 1, it becomes the steepest descent method with a smaller learning rate. In the iterative process, μ is adaptive. If the training is successful, the value of μ will be reduced. If it fails, it increases the value of μ and eventually reduces the training function to a certain value. Generally, the mean-variance of the function is less than 0.001.

5 Classification system based on fuzzy rule

FRBCSs are helpful and well-established application in machine learning plat from since they provide an interpretable method for end users’. This model of classification has two vital parts, the knowledge base is association of a rule base, associated by a group of fuzzy based rule and the data base that gathers the membership functions of the fuzzy sections determined to the input variables. If knowledge obtained by experts of the issue is not found, it is necessary for using certain process involving machine learning to get the knowledge base.

Any issues of classification including of M training patterns a_l = (a_i1 … a_lx), l = 1, 2 ... x from N classes where a_lj is the jth value (j = 1,2, ... , x) of the lth pattern of training.

In this model, we utilize fuzzy rules of the below given form for modelling classifier:

Rule F_I: If a₁ is $\hat{B_{1}^{i}}$ and …. And a_x is $\hat{B_{x}^{i}}$ then class= D_i with

$F V_{I} .$ (18)

Here F_i is ith class label, dimensional pattern is given as a = (a₁, …… , a_x), $\hat{B_{x}^{i}}$ is a group linguistic fuzzy. We utilize triangular membership function as a label of linguistic form whose association will formed an antecedent fuzzy set. This type of rule is called a DNF fuzzy rule.

For estimation of weight of rule, various heuristic have been implemented. In this model, we include Fhe rule weight as confidence of fuzzy or Certainty factor (CF) as shown in Eqn (19). ${FV}_{i} = {DL}_{i} = \frac{\sum a_{x} \in Class D_{i} μ \hat{B_{x}} (a_{x})}{\sum_{l = 1}^{M} μ \hat{B_{x}} (a_{x})}$ (19)

6 Classification method for nearby rock stability of coal roadway

6.1 The selection of classification index

On basis of initializing the factor influencing the rock stability, the lower input and output details is formed as index for classification of nearby rocks.

(1) Input information

A: roof strength; B: coal seam strength; C: floor strength; D: buried depth of roadway; E: the ratio of the mining height; F: roadway pillar width; G: direct top first caving step.

(2) Output information

The quality grade of surrounding rock: the output layer results (00001), (00010), (00100), (01000), and (10000) respectively represent the quality grade of V, IV, III, II, I. Table 2 shows the stability type and the output information.

Table 2
Stability type and the output information

Stability type Output details

1 very stability 1 0 0 0 0

2 stability 0 1 0 0 0

3 relatively stability 0 0 1 0 0

4 instability 0 0 0 1 0

5 very instability 0 0 0 0 1

Stability type	Output details
1	very stability	1	0	0	0	0
2	stability	0	1	0	0	0
3	relatively stability	0	0	1	0	0
4	instability	0	0	0	1	0
5	very instability	0	0	0	0	1

6.2 Design of the model

An improved BP-ANN was used to establish a model for nearby rock classification stability. The model is implemented by three seams, mentioning the input seam, the hidden layer, and the output seam. The input layer is the classification index of the stability found in the nearby rock. The hidden layer extracts and stores the nonlinear mapping relation in the sample data, and the hidden layer nodes are selected as 8. The output layer is the constancy type of nearby rock of coal.

6.3 Design of the sample

The input details of the component is composed of 7 input parameters, which associates to 7 classification indexes, and is expressed as the standardized value of each classification index. The specific values of stability of nearby rocks for the classification index of the coal roadway are arranged into the original data table. Then the original data table is standardized to obtain the standard sample.

Through the investigation of coal mines in many places, the data of 40 coal lanes were gathered and utilized to design training examples and test samples. The training samples include data of 35 coal lanes, which are used for training the model so that it can improve the classification accuracy by continuously learning and mastering the internal law of samples. The test samples include the data from 10 coal roadway to test the model and verify the applicability and accuracy of the classification procedure of nearby rock of coal roadway. Due to limited space, only the training samples containing 15 coal lanes are listed, as shown in Table 3.

Table 3
Training samples

No. Input information

A B C D E F G

1 0.12 0.83 0.51 0.02 0.93 0.97 0.15

2 0.25 0.01 0.32 0.02 0.97 0.93 0.03

3 0.44 0.49 0.58 0.13 0.15 0.14 0.29

4 0.07 0.67 0.45 0.57 0.02 0.99 0.55

5 0.41 0.52 0.55 0.35 0.08 0.96 0.11

6 0.28 0.01 0.32 0.04 0.98 0.14 0.07

7 0.58 0.90 0.29 0.13 0.03 0.07 0.69

8 0.12 0.56 0.98 0.74 0.06 0.28 0.45

9 0.14 0.56 0.96 0.69 0.07 0.38 0.48

10 0.28 0.98 0.76 0.12 0.44 0.99 0.35

11 0.15 0.67 0.98 0.73 0.5 0.92 0.44

12 0.05 0.89 0.30 0.04 0.66 0.99 0.19

13 0.07 0.88 0.29 0.02 0.99 0.15 0.15

14 0.06 0.67 0.73 0.99 0.05 0.95 0.20

15 0.44 0.55 0.59 0.14 0.15 0.94 0.27

No.	Input information
1	0.12	0.83	0.51	0.02	0.93	0.97	0.15
2	0.25	0.01	0.32	0.02	0.97	0.93	0.03
3	0.44	0.49	0.58	0.13	0.15	0.14	0.29
4	0.07	0.67	0.45	0.57	0.02	0.99	0.55
5	0.41	0.52	0.55	0.35	0.08	0.96	0.11
6	0.28	0.01	0.32	0.04	0.98	0.14	0.07
7	0.58	0.90	0.29	0.13	0.03	0.07	0.69
8	0.12	0.56	0.98	0.74	0.06	0.28	0.45
9	0.14	0.56	0.96	0.69	0.07	0.38	0.48
10	0.28	0.98	0.76	0.12	0.44	0.99	0.35
11	0.15	0.67	0.98	0.73	0.5	0.92	0.44
12	0.05	0.89	0.30	0.04	0.66	0.99	0.19
13	0.07	0.88	0.29	0.02	0.99	0.15	0.15
14	0.06	0.67	0.73	0.99	0.05	0.95	0.20
15	0.44	0.55	0.59	0.14	0.15	0.94	0.27

The surrounding rock stability classification model using BP-ANN proposed in this paper is designed based on MATLAB, and the learning samples are used for learning and training. Through repeated training and learning, according to the actual output information of the model and the expected output information in the sample, the weights and thresholds of each layer of network are constantly revised and optimized. After 6543 cyclic calculation, the mean square error of the classification model is less than 0.01.

To test the applicability and accuracy of the classification stability of nearby rocks situated coal roadway using the BP-ANN proposed in this paper, and the stability of rock present in multiple coal roadway is classified and identified. The 13 input samples were used as input information, and the surrounding rock stability identification model of the coal roadway was used for type identification. The results were compared with the actual type of surrounding stability of rock of the coal runway, as shown in Table 4.

Table 4

Result of testing

No.	Result of output in testing					Results of identification	Actual type
1	0	0	0	1	0	IV	IV
2	0	1	0	0	0	II	II
3	1	0	0	0	0	I	I
4	0	0	1	0	0	III	III
5	0	0	0	0	1	V	V
6	1	0	0	0	0	I	I
7	0	0	0	1	0	IV	IV
8	1	0	0	0	0	I	I
9	0	0	0	0	1	V	V
10	1	0	0	0	0	I	II
11	0	0	0	0	1	V	V
12	0	0	0	1	0	IV	IV
13	0	0	1	0	0	III	III

As can be seen from Table 4, the identification accuracy of this model is 92.3%. This indicates that the nonlinear fitting effect of the classification model is good, and the complex mapping law between the stability for index in classification and stability type of nearby rock of coal roadway is mastered, which has good applicability and accuracy.

The difference in variation of R² for the proposed BP-ANN associated with the population utilized in BP-ANN is illustrated in Fig. 2, whereas the difference in variation of RMSE value for the proposed model with the used population size is depicted in Fig. 3. Figures 4 and 5 illustrates the real time and settlement values of the proposed BP-ANN values respectively.

Fig. 2

Variation of R² associated with size of population in BP-ANN.

Fig. 3

Variation of RMSE associated with size of population in BP-ANN.

Fig. 4

R² Value for proposed BP-ANN Model.

Fig. 5

R² Settlement Value for proposed BP-ANN Model.

7 Conclusion

Aiming at the classification of roadway surrounding rock stability, a classification method for finding the rocks nearby stability on basis of improved BP-ANN is proposed in this paper. To resolve the problem of lower speed convergence and easily to deploy into minimum of BP-ANN, an improved BP-ANN algorithm based on additional momentum and Levenberg-Marquardt optimization is proposed based on the analysis of the existing improved methods, which improves the convergence speed and avoids the local minimum effectively. Finally, the stability classification model of surrounding rocks of coal roadway using BP-ANN was established in MATLAB environment, and the model was applied to 13 data samples of coal roadway for testing, with the identification rate of 92.3%. The experimental results verify that the method proposed in this paper has a high accuracy of type identification and is applicable for finding the classification stability of rock found in nearby roadways.

References

Chen

T.Y.

and Li

C.H.

, Determining objective weights with intuitionistic fuzzy entropy measures: a comparative analysis, Information Sciences 180(21) (2010), 4207–4222.

Ding

, Gu

S.T.

, Guo

J.X.

, et al., Numerical Investigation on Factors Affecting Stability of Roadway Surrounding Rock with Fractured Roof, Geotechnical and Geological Engineering 37 (2019), 2373–2385.

Duan

B.F.

, Xia

H.L.

and Yang

X.X.

, Impacts of bench blasting vibration on the stability of the surrounding rock masses of roadways, Tunnelling and Underground Space Technology 71 (2018), 605–622.

Fang

and Shen

, Multi-attribute multi-scheme large group decision-making method about the option of slope treatment. Resource. Environ. Eng, 2011(4) (2011).

Huang

, Huang

, Ju

and Li

, Automated tunnel rock classification using rock engineering systems, Engineering Geology 156 (2013), 20–27.

Jato-Espino

, Castillo-Lopez

, Rodriguez-Hernandez

and Canteras-Jordana

J.C.

, A review of application of multi-criteria decision making methods in construction, Automation in Construction 45 (2014), 151–162.

Jiang

, Song

S.G.

, Li

, et al., Study on Surrounding Rock Stability of Small Clear-Distance Twin Highway Tunnel with Eight Lanes, Geotechnical and Geological Engineering 37 (2019), 593–598.

Khakestar

M.S.

, Hassani

, Moarefvand

and Madani

, Application of multi-criteria decision making methods in slope stability analysis of open pit mines, J Geol Soc India 87(2) (2016), 213–221.

Kong

X.S.

, Shan

R.L.

, Xiao

Y.H.

, et al., Classification of surrounding rock stability of coal roadways in fenxi coal mining area based on SPSS and ANN, Mining Safety & Environment Protection 43 (2016), 13–17.

10.

Lei

W.L.

, Zhou

H.L.

, Jiang

S.Y.

, et al., Application of Normal Cloud Model in Surrounding Rock Classification of Coal Roadway, Safety in Coal Mines 48 (2017), 94–197.

11.

D.Y.

, Li

X.B.

, Zhang

, et al., Stability Analysis of Surrounding Rock of Multi-arch Tunnel Based on Coupled Fluid-solid Theorem, Chinese Journal of Rock Mechanics and Engineering 26 (2007), 1056–1064.

12.

Liu

and Chen

, A new approach for application of rock mass classification on rock slope stability assessment, Eng Geol 89(1/2) (2007), 129–143.

13.

Liu

, Chen

and Zhou

, Stability evaluation of depth cutting slope in rock massbased on fuzzy probability method, Rock Soil Mech 29(S1) (2008), 368–372.

14.

Mikaeil

, Ozcelik

, Yousefi

, Ataei

and Hosseini

S.M.

, Ranking thesawability of ornamental stone using Fuzzy Delphi and multi-criteria decision-making techniques, Int J Rock Mech Min Sci 58 (2013), 118–126.

15.

Mirhossein

, Yarahmadi-Bafghi

A.R.

and Rouzbeh

G.M.

, Evaluation of the capability of key-groups method development using numerical methods forfindingthe critical key-group in stability analysis of rock slopes, Iranian J Min Eng 11(32) (2016), 23–32.

16.

Ramzan

, Gul

and Kadry

, Onset of Cattaneo-Christov heat flux and thermal stratification in ethylene-glycol based nanofluid flow containing carbon nanotubes in a rotating frame, IEEE Access 7, Pages 146190–146197.

17.

Prathik

, Anuradha

and Uma

, Prediction of Suitability of Soil for Different Crops using Spatial Data Mining, International Journal of Engineering and Advanced Technology 9(1) (2019), 2330–2337.

18.

Qiu

D.H.

, Chen

J.P.

, Que

J.S.

, et al., Evaluation of tunnel rock quality with routh sets theory and artificial neural networks, Journal of Jilin University (Earth Science Edition) 38(1) (2008), 86–91.

19.

Rafiee

, Ataei

, KhaloKakaie

, Jalali

S.M.E.

and Sereshki

, A fuzzy rockengineering system to assess rock mass cavability in block caving mines, 27(7) (2016), 2083–2094.

20.

Rizk-Allah

R.M.

, Hassanien

A.E.

and Elhoseny

, A Multi-Objective Transportation Model under Neutrosophic Environment, Computers and Electrical Engineering 69 (2018), 705–719.

21.

Shan

R.L.

, Kong

X.S.

and Shan

, Research progress of roadway surrounding rock stability classification based on neural networks, Mining Safety & Environment Protection 41 (2014), 109–112.

22.

, Li

, Cao

and Wen

, Macro-comprehensive evaluation method of highrock slope stability in hydropower projects, Stoch Environ Res Risk Assess 28(2) (2014), 213–224.

23.

Sudin

M.N.

, Abdullah

S.S.

and Nasudin

M.F.

, Humanoid Localization on Robocup Field using Corner Intersection and Geometric Distance Estimation, International Journal of Interactive Multimedia and Artificial Intelligence 5(Regular Issue) (2019), 50–56.

24.

Wang

M.W.

, Chen

G.Y.

and Jin

J.L.

, Risk evaluation of surrounding rock stability based on stochastic simulation of multi-element connection number and triangular fuzzy numbers, Chinese Journal of Geotechnical Engineering 33(4) (2011), 643–647.

25.

Wang

, Xu

, Li

, et al., A novel model of set pair analysis coupled with extenics for evaluation of surrounding rock stability, Mathematical Problems in Engineering, 2015; 1–9.

26.

Wang

, Sun

, Shang

, et al., Application of efficacy coefficient method to instability risk early-warning of tunnel surrounding rock, Chinese Journal of Rock Mechanics and Engineering 29(2) (2010), 3679–3684.

27.

M.J.

, Chen

Z.G.

, Xu

X.B.

, et al., The Present and Prospect of Highway Tunnel Surrounding Rock Stability, Journal of Chongqing Jiaotong University 22 (2003), 24–28.

28.

, Wang

and Zhang

, Stability analysis method of loess cutting slope basedon fuzzy set pair analysis, Rock Soil Mech J 2008(S1) (2008), U416.14.

29.

Xing

X.K.

, Zhang

K.P.

, Yan

M.L.

, et al., Model Test Research on Influence of Earth Tunnel Support Patterns on Stability of Surrounding Rock, Rock and Soil Mechanics 35 (2014), 2157–2162.

30.

Zhang

, Li

X.B.

and Gong

F.Q.

, Stability classification model of mine-lane surrounding rock based on distance discriminant analysis method, Journal of Central South University of Technology 15(1) (2008), 117–120.

Fuzzy rule based classification method of surrounding rock stability of coal roadway using artificial intelligence algorithm

Abstract

Keywords

1 Introduction

2 Related work

3 Influencing factors of surrounding rock stability of coal roadway

4.1 Principle of BP-ANN

6.1 The selection of classification index

Table 2 Stability type and the output information Stability type Output details 1 very stability 1 0 0 0 0 2 stability 0 1 0 0 0 3 relatively stability 0 0 1 0 0 4 instability 0 0 0 1 0 5 very instability 0 0 0 0 1

6.3 Design of the sample

References

Table 2
Stability type and the output information

Stability type Output details

1 very stability 1 0 0 0 0

2 stability 0 1 0 0 0

3 relatively stability 0 0 1 0 0

4 instability 0 0 0 1 0

5 very instability 0 0 0 0 1