 Open Access
 Total Downloads : 39
 Authors : Navdeep Kumar, Balmukund Mishra, Dr. Vikram Bali
 Paper ID : IJERTCONV5IS11008
 Volume & Issue : NCIETM – 2017 (Volume 5 – Issue 11)
 Published (First Online): 24042018
 ISSN (Online) : 22780181
 Publisher Name : IJERT
 License: This work is licensed under a Creative Commons Attribution 4.0 International License
A Novel Approach for BlastInduced Flyrock Prediction Based on Particle Swarm Optimization and Artificial Neural Network
Navdeep Kumar1, Balmukund Mishra2 DR. Vikram Bali3 1 Student , Computer Science Dept., P.I.E.T Samalkha 2Assistant Professor, Computer Science Dept., P.I.E.T Samalkha 3Assistant Professor, Computer Science Dept., P.I.E.T Samalkha
Abstract Flyrocks are the excessive rock fragments. There random throw from a blast can travel a large distances which may be beyond the blast safety area. This process of the blasting operation results in human injuries, fatalities, and structural damage. There are various empirical relationships which have been established to predict flyrock resulted from blasting. These practical methods only studypartial numbers of activefactors such as flyrock distance. But, the blasting also affected by other parameters such as blast geometry and geological conditions. Due to this disadvantage, the empirical methods lacks in accuracy, even in accuracy of the flyrock distance. In this research work, a method is proposed to predict the flyrocks. These rocks are made by blasting over a freshmethod. This approach is built on the mixture of Particle Swarm Optimization and Artificial Neural Network. Here ANN used to predict flyrock distance. Generally ANN used as one of the forceful areas of research in advanced and varied applications of science. ANN has the ability to right to map the input to output patterns. Also, it utilizes all influential parameters in case of prediction of flyrock distance. But, there are still some limitations concern to ANN
i.e. the rateof slow learning and getting stuck in local minima. PSO can be used to overcome these shortcomings. PSO is generally utilized in the various optimization engineering problems. This research work offerings a mix PSOANN predictive model for flyrock prediction. The results of the developed model are compared to the results of ICAANN, BPANN, empirical equation and multivariate regression analysis (MRA). The parameters for comparison are (Root Mean Square Error), Coefficient of Determination (R^2) and Least Cost. These parameters are firstly calculated by comparing testing and trained data from ANN. These parameters are than compared with that of existing methods
i.e. ICAANN, BPANN, empirical equation and multivariate regression analysis (MRA). MATLAB R2013a is used as an implementation platform using general MATLAB toolbox and Artificial Neural Network toolbox.
Keywords: Artificial neural network, Imperialist competitive alg orithm, flyrock, Blasting etc.
I.INTRODUCTION
One of the crucial components of the surface mining is blasting. It serves as the most important role in dividing [6] burden and disclosing coal and inother mineral deposits [1].The boundaries of the blast areaare determined by the blaster and the flyrock is not probable to travel outside the blast zone. For the period of blasting, all employees must be detached from the blast part. Also all entries to the blast area
must be guarded. If someone is essential to visitclassified the blast area, a proper blasting safetyshould be taken. [11].

FLYROCK AND ITS PARAMETERS
Flyrock is unnecessary rock garbage thrown during bench blasting in mines [1]. It is driven rock fragments by energy of explosive away from the blast zone. It is objectionable environmental effects of blasting operations. In this there is an affective relationship among explosive energy sharing, rock mass automatic strength, and charge limitation.

IMPERIALIST COMPETITION ALGORITHM
It isnewoverall search [9] experiential that uses colonization [2] and imperialistic competition process and used expansively to solve different types of optimization problems [4]. This algorithm starts with some early countries. It is new evolutionary algorithm that is moved by the humans sociopolitical progress [8]. Each singular of the population is called country. Population is divided in two parts, colonies and imperialist state. The competition between imperialists to take ownership of the colonies of each other forms this algorithm. In this competition the weak empires collapse slowly and finally one imperialist and other country is its colony [3].After dividing all colonies among imperialists and creating the initial empires, these colonies move to their important imperialist.
This movement is simple
model of assimilation policy that was assumed.
Figure 1 Movement of colonies to their related imperialist [3].
In ICA the optimization process starts with producing the population. This algorithm, each unit of the population called a country. The countries are distributed in two sections; the finest countries are considered to be imperialists and the rest of the countries form the colonies. All colonies are distributed between the remaining imperialists on the basis power. Combination of each imperialist organized with its colonies makes an empire. After the initial empires, the colonies move to relevant imperialists and keep the assimilation of imperialist states. The following steps ICA optimization procedure is [1].

Initial Empires optimization procedure
Starts with initializing the entities section which are called countries. In this problem, a country have 1*N variables range. This array is defined as follows:
Country = [A1 , A2 , A3 , . . . , A variable]
In thisthe particle which need to best solution. In a country, each parameter can be considered as a human related characteristic such as culture and Language, in which this makes an attempt to find the best combination of thesecharacteristics. Cost function country is as follows:
F (Country) = [A1, A2, A3, . . . , A variable]
The technique of ICA optimization starts with size of countries, country and select a powerful countries as the ( imperialist), remaining of the countries are measured as a colonies ( colony). The colonies are distributed into imperialists based on power to makingoriginal empires. Therefore the normalized cost of each imperialist is defined as follows:
= max()
Inthis is cost of nth imperialist and is its normalized cost. The normalized power of each imperialist is as follows:
=
=1
The number of initial colonies for each empire is as follow: N.Cn= round {colony}
In which N.Cn is the initial colonies of thEmpire and
colony is the total number of initial colonies.
To distribute the colonies within imperialists, N.C.n of the colonies is accidentally selected and making to the th imperialist and therefore produce the nth empire.
A.Assimilation, Revolution, and Uniting
In this step, assimilation and revolution are the process. Assimilation is movement of colonies toward the imperialists where imperialists attempt to accept their colonies and makes part of them. This process is simulated by affecting all colonies to the imperialist along different axis.
Revolution is defined as changes in the power and structure that happen quickly. In optimization process, revolution makes sudden changes in sociopolitical things of a country. This action increases the optimal part of algorithm and makes quick result of countries to local minima.
Unitealike empires when distance between two imperialist becomes minus than the threshold distance. In this
scenario these imperialists are united and a new empire will be formed.

Imperialistic Competition
In ICA optimization procedure, all empires make an attempt th colonies of other empires. In this terminology this action is called imperialistic competition which is the final optimization step..
The imperialistic competition is shown in Figure 2
Figure 2: Imperialistic competition [1].
The virtual code of Imperialist competitive algorithm includes following steps.

Selection of unplanned points on the function and initialization of the empires.

Move the colonies to relevant imperialist. It is called Assimilation.

Unplanned change in position of colonies called revolution.

If there is colony in empire which has fewer cost than the imperialist, exchange the places of that colony and the imperialist and unites the alike empires.

Add the total cost of all empires.

Chose the weakest colony from weakest empires and provide it to one of the powerful empires and this is called Imperialistic competition.

Remove lowest weak empires, if break conditions fulfilled stop, if not go to 2.

ARTIFICIAL NEURAL NETWORK
It is a mathematical model that works on the basis of simulating the human brain. In other words, an ANN is nonlinear function approximation which comprehends a relationship between desired input data and output data. ANNs require training to learn and accordingly map a relationship from the data. The capability of ANNs that learn the samples and increase performance over. Learning is the property that makes ANN dissimilarto other networks. This ability comes from training algorithm [1].
ANNmethod to calculate blastinduced fly rock the pattern of the result is projected by ANN on the basis of preceding learning. Once the neural network has been trained, any similarities in the new pattern will be detected and the
result will change accordingly, thus providing the technique interpolation capability. ANN trained using back propagation algorithm. The feed forward BPNN back propagation comprises 3 layers, i.e. input layer visible, hidden layer not visible and output layer that gives result. Layers are made up by neurons i.e. the basic processing units. These neurons connect the layer using appropriate weight.
The output of the neurons in the input layer becomes input for the neurons in the hidden layer and the same scenario applies to connection between hidden and output layers. The problem defines the number of hidden layers which is not visible and the neurons in them. In the present case, algorithm and logsigmoid transfer function has been undertaken. After trial of a number of different and same combinations, two hidden layer between input output and ten neurons in each hidden layer have been found as the best model for the case under the study.
Here below the full steps for implementations:

Reading and inputting of data

Extraction of Last column (Y) and rest of the data (X) separately

Calculation of number of Rows and columns of extracted data

Normalization of extracted data

Finding of minimum and maximum value from the extracted matrix X

Finding minimum and maximum value from the extracted matrix Y

Declaration of a loop according to number of column of X

Normalization of X matrix column wise

Normalization of X matrix column wise

Generation of feedforward backpropagation network using number of rows of Xmatrix

Training of network using PSO method

Extraction of all the elements of network one by one

Computation of total number of elements in the network

Creation of ones matrix according to total number of elements

Inputting of PSO Algorithm's Parameters i.e. Size of swarm and maximum number of iterations, Cognition Coefficient Social Coefficient

Generation of Initial Population according to size of swarm and computation of best position and best cost of particle using ANN

Optimization of cost and position of the particles at each iteration

Updation of particle Velocity using Cognition Coefficient and Social Coefficient

Updation of position using updated velocity of particle

Updation of the cost using updated position of the particle and ANN

Final updation of the Position and cost of particle

Display and accumulation of best cost at each iteration

Plotting of all the best cost


Simulation of Trained Network using Testing and Training data matrix and getting of Testing and Training simulated optimized object

Calculation of Mean Square Error by comparison of optimized object matrix with initial Y matrix Testing and Training objects

Display of initial and final optimized training data

Display of initial and final optimized testing data

Display of Coefficient of determination (2) for Training data

Display of Coefficient of determination (2) for Testing data


EXPERIMENTAL RESULTS
platform using general MATLAB toolbox and Artificial Neural Network toolbox. In this research work, a method is proposed to predict the flyrocks. These rocks are made by blasting complete a newmethod. This approach is based on the combination of Particle Swarm Optimization (PSO) and Artificial Neural Network (ANN). Here ANN is used to predict flyrock distance. Generally ANN is used as one of the most forceful areas of research in advanced applications of engineering. ANN has the ability to right map input to output patterns. Also, it utilizes all influential parameters in case of prediction of flyrock distance. But, there are still some limitations concern to ANN i.e. the measuredspeed of learning and getting stuck in limitedjots. PSO can be used to overcome these shortcomings. PSO is generally utilized in the various optimization engineering problems. This research work presents a hybrid PSOANN predictive model for flyrock prediction. The results of the developed model are compared to the results of ICAANN, BPANN, empirical equation and multivariate regression analysis (MRA). The parameters for comparison i.e. Root Mean Square Error, Coefficient of Determination and Minimum Cost. These parameters are firstly calculated by comparing testing and trained data from ANN. These parameters are than compared with that of existing methods i.e. ICA ANN, BPANN, empirical equation and multivariate regression analysis (MRA). The value of these parameters has been given in Table 1. Table 1 Comparison of RMSE and Minimum Cost.
Method/Para meters
RMSE (Root Mean Square
Error)
Coefficient of Determination
(2)
Minimum Cost
PSOANN
(Proposed)
0.0393
0.9927
0.0030
ICAANN
6.582
0.981
0.067
BPANN
13.478
0.919
NA
MRA
23.877
0.743
NA
Empirical
109.064
0.118
NA
We have also given snapshots of some graphs and bar chart for showing the performance of proposed predictive model. Figure 3 is the snapshot of cost value w.r.t. number of repetitions. Its clearly showing that cost is exponentially decreasing w.r.t. number of iterations. Figure 4 is the snapshot of comparison of ANN output and actual input of training data w.r.t.
Number of nodes in unseen layer. Figure 5 is the snapshot of ANN output and actual input of testing data w.r.t. Number of nodes in hidden layer. Figure 6 is the snapshot of measured flyrock distance vs. Predicted flyrock distance in meters according to ANN testing data. Figure7 is the snapshot of measured flyrock distance vs. Predicted flyrock distance in meters according to ANN training data. Figure 8 is the snapshot of Bar chart showing comparison of RMSE for ICAANN and PSOANN. Figure 9 is the snapshot of Bar chart showing comparison of Coefficient of determination (R^2) for Testing data for ICAANN and PSOANN. Figure 10 is the snapshot of Bar chart showing comparison of minimum cost value for ICAANN and PSOANN.
Figure 3 snapshot of cost value w.r.t. number of iterations. Its clearly showing that cost is exponentially decreasing w.r.t. number of repetitions
Figure 4 snapshot of comparison of ANNoutputand actual input of training data w.r.t. Number of nodes in hidden layer
Figure 5 snapshot of ANN output and actual input of testing data w.r.t.
Number of nodes in unseen layer
Figure 6 snapshot of measured flyrock distance vs. Predicted flyrock distance in meters according to ANN testing data
Figure 7 snapshot of measured flyrock distance vs. Predicted flyrock distance in meters according to ANN training data
Figure 8 snapshot of Bar chart showing comparison of RMSE for ICA ANN and PSOANN
Figure 9 snapshot of Bar chart showing comparison ofCoefficient of determination (R^2) for Testing data for ICAANN and PSOANN
Figure 10 snapshot of Bar chart showing comparison of minimum cost value for ICAANN and PSOANN

CONCLUSION
A predictive model based on the arrangement of PSO and ANN is developed to predict flyrock which is made by blasting. An exactly recorded and collected data is utilized to train the PSOANN predictive model. Hole depth, load to spacing ratio, reducing length, burden per delay, powder factor, rock density considered as input limitations. Fly rock distances are assigned as the output parameter. It can be concluded from the experimental results that the proposed model is well able to expect flyrock distance with high mark of correctness. The proof of previous statement is the snapshots of measured flyrocks for testing and training data in last chapter. Measured flyrocks are very much closer to that of predictive flyrocks. Also, for comparison purpose, the results of proposed method are compared with existing methods such as ICAANN, BP ANN, empirical equation and multivariate regression analysis (MRA). Also, it is surveyed that the existing predictors provide very quick and simple prediction, whereas the proposed PSOANN model exhibited higher prediction performance model compared to other methods.
REFERENCES

Marto, Aminaton, et al. "A novel approach for blastinduced flyrock prediction based on imperialist competitive algorithm and artificial neural network." The Scientific World Journal 2014 (2014).

Ghanavati, Mojgan, et al. "An Efficient Cost Function for Imperialist Competitive Algorithm to Find Best Clusters." Journal of Theoretical & Applied Information Technology, 29.1 (2011).

SanazAsfia, ArashGhorbanniaDelavar The proposed Center Initialization Based on Imperialist Competitive Algorithm (CIB ICA) Journal of mathematics and computer science 10 (2014), 297310.

Niknam, Taher, et al. "A new hybrid imperialist competitive algorithm on data clustering." Sadhana 36.3 (2011): 293315.

Ghanavati, Mojgan, et al. "Hybrid Imperialist Competitive Algorithm and Dynamic Validity Index to find the best clusters." May 1217, 2011.

Armaghani, DanialJahed, et al. "Application of two intelligent systems in predicting environmental impacts of quarry blasting." Arabian Journal of Geosciences 8.11 (2015): 96479665.

Trivedi, Ratnesh, et al. "Application of Artificial Neural Network for Blast Performance Evaluation." International Journal of Research in Engineering and Technology. Volume: 03 Issue: 05, May2014. Pp. 564574.

Maadi, Marjan, and MasourehMaadi. "Optimization of Cluster Heads Selection by Imperialist Competitive Algorithm in Wireless Sensor Networks." International Journal of Computer Applications 89.19 (2014): 2934.

Karami, S., and Sh B. Shokouhi. "Application of imperialist competitive algorithm for automated classification of remote sensing images."International Journal of Computer Theory and Engineering 4.2 (2012): 137.

Trivedi, Ratnesh, T. N. Singh, and Neel Gupta. "Prediction of blast induced flyrock in opencast mines using ANN and ANFIS." Geotechnical and Geological Engineering 33.4 (2015): 875891.

Bajpayee, T., H. Verakis, and T. Lobb. "An analysis and prevention of flyrock accidents in surface blasting operations." Proceedings of the annual conference on explosives and blasting technique. Vol. 2. ISEE; 1999, 2004.

Rehak, T., et al. "Flyrock issues in blasting." Proceedings of the annual conference on explosives and blasting technique. Vol. 1. ISEE; 1999, 2001.

ZHOU, Zilong, et al. "Safety Evaluation of Blasting Flyrock Risk with FTA Method." Pp.11841187.