Forecasting The Production of Fully-Mechanised Coal Face by Meta-Synthetic AI Method

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Forecasting the Production of Fully-Mechanised Coal Face by

Meta-Synthetic AI Method
Han Wanlin1 and Zhang Youdi2

ABSTRACT reduce trying error and reach the request. Thereby, experts’
Being analysed systematically, the factors influencing the production of
experience and knowledge which is involved in samples and
fully-mechanised coal face have been divided into geological, mining hard to express by structured type are distributed in the
technical, and management groups. Artificial neural network-expert network’s linking matrix; studying and training samples are
systems (ANN-ES) are established to decide the influencing factors’ remembered by adjusting linking weights and node threshold;
membership grade; genetic algorithm-artificial neural network and the relation between input variables and output variables is
(GA-ANN) are used to decide the weights; fuzzy comprehensive ascertained.
evaluation method is used to calculate the integration indexes. Finally, the
Knowledge acquirement and denotation are the premise of
GA-ANN model is established to forecast the production of fully-
mechanised coal face. It is proved that the model is reliable with higher knowledge reasoning. Knowledge acquirement includes:
precision. 1. Fixing on neural network’s type and structure factors.
The production of fully-mechanised coal face is influenced by many
factors. However, some influencing factors are neglected by the 2. Confirming network’s learning factors.
traditional forecasting methods, the results of which are not precise and 3. Organising studying samples set to be trained.
hardly be used as a guide for mine production and management. The
relation between the production of fully-mechanised coal face and its 4. Network is used to study samples to get weights
effecting factors is analysed, and the meta-synthetic AI approaches are distribution. Experts’ experience and knowledge involved in
used to forecast the production of fully-mechanised coal face. samples are distributed in the network’s linking weight
matrix and node’s threshold vector.
SOME META-SYNTHETIC AI APPROACHES
Genetic Algorithm-Artificial Neural Network
The conception of meta-synthetic AI technique should include
the following aspects (Zhang, Han and Li, 2002; Z Youdi and (GA-ANN)
Wanlin, 2001): Training forward neural network is an optimal problem to find
1. the meta-synthesis of digital information and knowledge the best structure and linking weight to make the difference
information; between the output of neural network and the goal function least.
Therefore, genetic algorithm can be used to train forward neural
2. the meta-synthesis from qualitative to quantitative analysis; network. According to the character of neural network, training
3. the meta-synthesis of exact and inexact processing; network by GA is different from traditional genetic algorithm as
follows (Wang Yaonan, 1996).
4. the meta-synthesis among different AI branches; and
5. the comprehensive application of AI branches and other Coding
optimisation techniques.
1. Coding of linking weights
Artificial neural network-expert system (ANN-ES) Each unit of genetic algorithm for artificial neural network is an
individual network, and the referred factors as linking weights
As an important and active branch of AI, expert system has between nodes are all real numbers. Therefore real number
drawbacks such as bottleneck in obtaining knowledge, coding is adopted here.
narrowness of knowledge domain, low efficiency of reasoning,
bad adapting ability, and lack of hierarchy. Furthermore, expert 2. Expression of the space frame
system cannot imitate reasoning process of human expert. These Hidden neuron is used as basic genotype, in which each hidden
restrict the sustaining power of expert system. Artificial neural neuron takes part in evolution with its linking weight between
network is combined with expert system to form the neural input layer and output layer as a whole. Compared with the
network-expert system (ANN-ES) in which expert system may traditional binary system string, coding of each individual seems
stand for intelligence’s cognition and artificial neural network as a queue with hidden neuron package. This coding type can
reflects intelligence’s perception (Lu Weimin, 1998). gain neural network structure conveniently.
Knowledge explanation of ANN-ES is completely different
from traditional AI. It is a hidden denotation, in which the The three basic operators
system masters knowledge by using examples and their solutions
offered by experts. As using real number coding type, genetic algorithm for neural
network is different from standard genetic algorithm. A simple
The examples are studied through given algorithm.
discussion is as follows.
Distribution of network’s linking weights and thresholds is
modified continually by self-adapting algorithm in the system to 1. Reproduction
Common roulette is adopted in reproduction, which is Monte
Carlo probability selection based on the fitness. Preeminent
1. Tongji University, Shanghai 200092, PR China. selection enters at the same time, which can keep the best
2. China University of Mining and Technology, Xuzhou 221008, individual of each generation down to filial generation.
PR China.

Mine Planning and Equipment Selection Kalgoorlie, WA, 23 - 25 April 2003 117
HAN WANLIN and ZHANG YOUDI

2. Crossover Management factors


Crossover in GA-ANN is defined as the course in which Management factors reflect the management level and worker’s
individuals of neural network exchange their hidden neurons. capability of the fully-mechanised coal face, which include
Different from the fixing point crossover of genetic algorithm, degree of reaching design production rate, mining labour
the crossing points of the two individuals which took part in productivity, mining equipment availability, safety factor, mining
crossover can be different, and the number of genes, ie the machine utilisation and prop consuming.
number of hidden neurons they exchange can also be different,
which makes the number of genes namely the network size QUANTIFYING OF FULLY-MECHANISED
change. It carries out studying on network structure as well as its
weights. COAL FACE’S EFFECTING FACTORS
3. Mutation ANN-ES to decide the effecting factors’
A trend algorithm is adopted. Its keystone is to add white yawp membership grade
disturbing with zero mean value to the linking weights and check
the network. If the new weights can receive smaller error, the According to the principle of ANN-ES, 15 ANN-ES are
result is accepted, otherwise the result is cancelled and new established to decide the effecting factors’ membership grade of
fully-mechanised coal face’s production. The 15 ANN-ES are all
disturbing is acceded to. Certainly BP algorithm can also be used
three-layer BP neural networks, in which the input layer is
to study the weights.
quantified value of the effecting factors, the output layer is their
membership grade. The existed fruit of constructing membership
FACTORS EFFECTING THE PRODUCTION OF grade and the adjusted results of membership grade by experts
FULLY-MECHANISED COAL FACE based on technique progress are used comprehensively to train
the ANN-ES. Membership grades of the effecting factors which
The factors influencing the production of fully-mechanised coal are implicated in samples and hard to express by structured type
face have been divided into geological, mining technical, and are distributed in the network’s linking matrix, studying and
management groups (Zhang Xianchen and Zhu Jianmin, 1991; training samples are remembered by adjusting linking weights
Xing Cunen and Song Yi, 1997). The three groups construct a and node threshold, and the relation between input variables and
uniform entity, which are shown in Table 1. output variables are obtained.

Geological factors GA-ANN to decide effecting factors’ weights


There are many geological factors influencing the production of In order to estimate each effecting factor’s effect on fully-
fully-mechanised coal face. According to their characters, seven mechanised coal face’s production, GA-ANN is adopted to
factors are chosen, they are: complex degree of geological decide their weights. Three three-layer GA-ANN are established:
structure, coal seam stability, coal seam thickness, coal seam dip the first GA-ANN is geological GA-ANN, the seven factors’
angle, coal seam hardness, top and bottom conditions of coal membership grade are input nodes, their appraised values are
seam and the coal face’s advancing length. output nodes; the second GA-ANN is mining technical
GA-ANN, the two factors’ membership grades are input nodes,
Mining technical factors their appraised values are output nodes; the third GA-ANN is
management GA-ANN, the six factors’ membership grades are
Mining technical factors effect the production of fully- input nodes, their appraised values are output nodes. According
mechanised coal face directly. Coal face length and integrated to the principle of GA-ANN, genetic algorithm is used to train
matching adaptability of the mining equipment in the coal face the neural network to gain its hidden neuron’s number and
are chosen. weights.

TABLE 1
The structure of factors influencing the production of fully-mechanised coal face

Complex degree of geological structure


Coal seam stability
Coal seam thickness
Geological factors Coal seam dip angle
Coal seam hardness
Top and bottom conditions of coal seam
Coal face’s advancing length
Production of fully-mechanised
Coal face’s length
coal face Mining technical factors
Integrated matching adaptability of the mining equipment in the coal face
Degree of reaching design production rate
Mining labour productivity
Mining equipment availability
Management factors
Safety factor
Mining machine utilisation
Prop consuming

118 Kalgoorlie, WA, 23 - 25 April 2003 Mine Planning and Equipment Selection
FORECASTING THE PRODUCTION OF FULLY-MECHANISED COAL FACE BY META-SYNTHETIC AI METHOD

Fifteen fully-mechanised coal faces’ condition of one mine are PRODUCTION FORECAST OF FULLY-
used as samples to train the three GA-ANN to get their weights
and threshold among neurons of each layer. According to the
MECHANISED COAL FACE BY GA-ANN
principle of deciding weights by neural network (Tan Zhanlu,
Xiong Guohua and Li Xiangyi, 1996), the weights among Design of GA-ANN
neurons of each layer of the three GA-ANN are used to decide Three-layer GA-ANN is adopted: the input nodes are the
the effecting factors’ weights, which are shown in Table 2. integrated indexes of geological, mining technical and
management groups; the output node is coal face’s production;
Compute of the effecting factors’ integrated the hidden neuron’s number and weights are decided by genetic
indexes algorithm.
Fuzzy comprehensive evaluation method is adopted to compute
the three effecting groups’ integrated indexes. The integrated Training samples’ selection and training of
index is: GA-ANN
Xi = ∑ U ij A ij Twelve fully-mechanised coal faces of the mine are used as
j samples to train the GA-ANN, which are shown in Table 3. The
where: maximum training step of the GA-ANN is ordered 1000, and the
Xi is ith integrated index, i = 1, 2, 3 allowed error is 0.001. The GA-ANN reached stable state after
87 steps, and the system error is 0.00023.
X1 is geological integrated index
Three fully-mechanised coal face’s productions of the mine are
X2 is mining technical integrated index forecasted by the trained GA-ANN, whose results are shown in
X3 is management integrated index Table 4. The error did not exceed five per cent. Therefore,
forecasting the production of fully-mechanised coal face by
Uij is the index membership grade of jth factor of ith group meta-synthetic AI methods has higher precision and is an
Aij is the weight of jth factor of ith group’s effect on fully- effective forecasting method.
mechanised coal faces’ production

TABLE 2
The effecting factors’ weights of fully-mechanised coal face’s production
Effecting factor Weight Effecting factor Weight
Complex degree of geological structure 0.231572 Coal face’s length 0.521625
Coal seam stability 0.217635 Integrated matching adaptability of the mining 0.478375
equipment in the coal face
Coal seam thickness 0.087846 Degree of reaching design production rate 0.178476
Coal seam dip angle 0.110297 Mining labour productivity 0.183055
Coal seam hardness 0.064853 Mining equipment availability 0.178997
Top and bottom conditions of coal seam 0.165193 Safety factor 0.122823
Coal face’s advancing length 0.122604 Mining machine utilisation 0.168941
Prop consuming 0.167708

TABLE 3
Training samples of the GA-ANN.

Coal face Production Geological integrated Mining technical integrated Management integrated
number (t/month) index index index
1 85 185 0.797345 0.817269 0.765389
2 123 660 0.783562 0.853126 0.786548
3 96 985 0.764131 0.834521 0.773689
4 166 207 0.745447 0.914321 0.823476
5 114 400 0.767681 0.824318 0.746532
6 121 480 0.769138 0.845314 0.788321
7 116 000 0.798352 0.867415 0.816753
8 101 800 0.822727 0.796546 0.795437
9 90 000 0.776119 0.824172 0.719863
10 140 000 0.794993 0.921163 0.843782
11 90 830 0.845629 0.756543 0.698731
12 90 000 0.823986 0.786542 0.734278

Mine Planning and Equipment Selection Kalgoorlie, WA, 23 - 25 April 2003 119
HAN WANLIN and ZHANG YOUDI

TABLE 4
The forecasting production and error by GA-ANN.

Coal face Geological Mining technical Management Real production Forecasting production Error
number integrated index integrated index integrated index value (t/month) value (t/month) (%)
1 0.817423 0.882543 0.836518 129 933 125 200 -3.64
2 0.838521 0.832194 0.736529 100 000 104 210 4.21
3 0.786515 0.865389 0.817698 110 000 104 700 -4.86

REFERENCES Zhang Xianchen, Zhu Jianmin, 1991. Evaluation on the geological


conditions in gently inclined or inclined coal seams from the point of
Lu Weimin, 1998. AI Technology and its Application, (Tongji University view of mining technology, Journal of China University of Mining
Publisher: Shanghai). and Technology, 20(3):35-44.
Tan Zhanlu, Xion, Guohua and Li Xiangyi, 1996. ANN-based factor Zhang, Y D and Han, W L, 2001. Mineral resource prediction by
analysis method and its application in mining, Metal Mine, integrated AI method, Tenth ISMPES, New Delhi, India.
12(3-5):36. Zhang, Y D, Han, W L and Li, X C, 2002. Application of meta-synthetic
Wang Yaonan, 1996. Intelligent Controlling System, (Hunan University AI technique for mining conditions evaluation, in Proceedings 30th
Publisher: Changsha). APCOM Symposium, Phoenix, USA.
Xing Cunen and Song Yi, 1997. A forecasting model on the production
of high-output and high-efficiency fully-mechanised coal face,
Shanxi Mining Institute Learned Journal, 15(3):268-275.

120 Kalgoorlie, WA, 23 - 25 April 2003 Mine Planning and Equipment Selection

You might also like