Professional Documents
Culture Documents
Download textbook Advances In Intelligent Systems And Interactive Applications Proceedings Of The 2Nd International Conference On Intelligent And Interactive Systems And Applications Iisa2017 1St Edition Fatos Xhafa ebook all chapter pdf
Download textbook Advances In Intelligent Systems And Interactive Applications Proceedings Of The 2Nd International Conference On Intelligent And Interactive Systems And Applications Iisa2017 1St Edition Fatos Xhafa ebook all chapter pdf
https://textbookfull.com/product/smart-sensors-networks-
communication-technologies-and-intelligent-applications-a-volume-
in-intelligent-data-centric-systems-fatos-xhafa/
https://textbookfull.com/product/intelligent-systems-and-
applications-proceedings-of-the-2020-intelligent-systems-
conference-intellisys-volume-2-kohei-arai/
https://textbookfull.com/product/intelligent-systems-and-
applications-proceedings-of-the-2020-intelligent-systems-
conference-intellisys-volume-3-kohei-arai/
International Proceedings on Advances in Soft Computing
Intelligent Systems and Applications ASISA 2016 1st
Edition M. Sreenivasa Reddy
https://textbookfull.com/product/international-proceedings-on-
advances-in-soft-computing-intelligent-systems-and-applications-
asisa-2016-1st-edition-m-sreenivasa-reddy/
https://textbookfull.com/product/intelligent-interactive-
multimedia-systems-and-services-2016-1st-edition-giuseppe-de-
pietro/
Fatos Xhafa
Srikanta Patnaik
Albert Y. Zomaya Editors
Advances in
Intelligent Systems
and Interactive
Applications
Proceedings of the 2nd International
Conference on Intelligent and
Interactive Systems and Applications
(IISA2017)
Advances in Intelligent Systems and Computing
Volume 686
Series editor
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
e-mail: kacprzyk@ibspan.waw.pl
About this Series
The series “Advances in Intelligent Systems and Computing” contains publications on theory,
applications, and design methods of Intelligent Systems and Intelligent Computing. Virtually
all disciplines such as engineering, natural sciences, computer and information science, ICT,
economics, business, e-commerce, environment, healthcare, life science are covered. The list
of topics spans all the areas of modern intelligent systems and computing.
The publications within “Advances in Intelligent Systems and Computing” are primarily
textbooks and proceedings of important conferences, symposia and congresses. They cover
significant recent developments in the field, both of a foundational and applicable character.
An important characteristic feature of the series is the short publication time and world-wide
distribution. This permits a rapid and broad dissemination of research results.
Advisory Board
Chairman
Nikhil R. Pal, Indian Statistical Institute, Kolkata, India
e-mail: nikhil@isical.ac.in
Members
Rafael Bello Perez, Universidad Central “Marta Abreu” de Las Villas, Santa Clara, Cuba
e-mail: rbellop@uclv.edu.cu
Emilio S. Corchado, University of Salamanca, Salamanca, Spain
e-mail: escorchado@usal.es
Hani Hagras, University of Essex, Colchester, UK
e-mail: hani@essex.ac.uk
László T. Kóczy, Széchenyi István University, Győr, Hungary
e-mail: koczy@sze.hu
Vladik Kreinovich, University of Texas at El Paso, El Paso, USA
e-mail: vladik@utep.edu
Chin-Teng Lin, National Chiao Tung University, Hsinchu, Taiwan
e-mail: ctlin@mail.nctu.edu.tw
Jie Lu, University of Technology, Sydney, Australia
e-mail: Jie.Lu@uts.edu.au
Patricia Melin, Tijuana Institute of Technology, Tijuana, Mexico
e-mail: epmelin@hafsamx.org
Nadia Nedjah, State University of Rio de Janeiro, Rio de Janeiro, Brazil
e-mail: nadia@eng.uerj.br
Ngoc Thanh Nguyen, Wroclaw University of Technology, Wroclaw, Poland
e-mail: Ngoc-Thanh.Nguyen@pwr.edu.pl
Jun Wang, The Chinese University of Hong Kong, Shatin, Hong Kong
e-mail: jwang@mae.cuhk.edu.hk
Albert Y. Zomaya
Editors
Advances in Intelligent
Systems and Interactive
Applications
Proceedings of the 2nd International Conference
on Intelligent and Interactive Systems
and Applications (IISA2017)
123
Editors
Fatos Xhafa Albert Y. Zomaya
Technical University of Catalonia School of Information Technologies
Barcelona University of Sydney
Spain Sydney, NSW
Australia
Srikanta Patnaik
Department of Computer Science
and Engineering, Faculty of Engineering
and Technology
SOA University
Bhubaneswar, Odisha
India
As we all know, Intelligent-interactive Systems (IIS) are systems that interact with
human beings, media, or virtual agents in intelligent computing environments. The
emergence of Big Data and Internet of Things opened new opportunities in both
academic and industrial research for successful design and development of
Intelligent-interactive Systems. These systems are usually embodied with human
capabilities such as to reason, learn, perceive, take decisions, make plans, and
perform actions in a wide range of applications such as smart navigation of auto-
mobiles so as to avoid traffic congestion, intelligent transportations, smart light,
virtual environments, mood-based music systems, and many more. The develop-
ment of these Intelligent-interactive Systems involves fabricating and controlling
machines to perform tasks usually need to be done by human beings.
Since research on Intelligent-interactive Systems deals with the pressing need of
designing innovative systems requiring new approaches to current computing
architectures and interactions between systems and their peripherals, it explicitly
demands knowledge acquisition in many cross-disciplinary subjects such as algo-
rithmic foundations, computational and mathematical modeling fundamentals,
artificial and cognitive intelligence foundations, human–computer interaction,
software architecture, and developmental concepts and logics of embedded systems.
Some of the significant underlying questions include the following: How huge
amounts of data are to be collected and how can they be further analyzed for
enhancing the interaction between intelligent systems and users, what are the
various issues and challenges faced during technical design and realization of
Intelligent-interactive Systems?
This volume explores how novel interactive systems can intelligently face var-
ious challenges and limitations previously encountered by human beings using
different machine learning algorithms along with analysis of recent trends.
However, designing and development of Intelligent-interactive Systems are quite
hard to implement due to variations in abilities of human being, preferences, and
limitations. The current researchers are focusing on two major areas: artificial
intelligence and human–computer interaction. Further, these two areas are classified
into application-specific sub-areas such as computer vision, speech recognition or
v
vi Preface
(IV) Mobile and Wireless Communication: This area is one of the leading areas
of IIS, which covers ubiquitous or mobile computing and networking. This
section covers twenty-three (23) contributions.
(V) Internet and Cloud Computing: It is one of the essential areas of IIS, which
caters to enhance communication between the system and users, in a way
which may not be closely related to the system’s main function. This is
commonly found in the areas of multimodal interaction, natural language
processing, embodied conversational agents, computer graphics, and acces-
sible computing. In this section, there are four (4) contributions consisting of
micro-blogging, user satisfaction modeling to the design and construction of
graphical cloud computing platform.
(VI) Intelligent Systems: This is the nervous system of IIS, and many researchers
are engaged in this area of research. This section contains 11 contributions.
(VII) Applications: Applications of IIS in various domains are covered in the last
section which consists of 21 contributions.
Acknowledgements
The contributions covered in this proceeding are the outcome of the contributions
from more than hundred researchers. We are thankful to the authors and paper
contributors of this volume.
We are thankful to the editor in chief of the Springer Book series on “Advances
in Intelligent Systems and Computing” Prof Janusz Kacprzyk for his support to
bring out the second volume of the conference, i.e., IISA-2017. It is noteworthy to
mention here that constant support from the editor in chief and the members of the
publishing house makes the conference fruitful for the second edition.
We would like to appreciate the encouragement and support of Dr. Thomas
Ditzinger, executive editor, and his Springer publishing team.
We are thankful to our friend Prof. Madjid Tavana, Professor and Lindback
Distinguished Chair of Business Analytics of La Salle University, USA, for his
keynote address. We are also thankful to the experts and reviewers who have
worked for this volume despite the veil of their anonymity.
We look forward to your valued contribution and supports to next editions of the
International Conference on Intelligent and Interactive Systems and Applications.
We are sure that the readers shall get immense benefit and knowledge from the
second volume of the area of Intelligent and Interactive Systems and Applications.
Fatos Xhafa
Srikanta Patnaik
Albert Y. Zomaya
Contents
Autonomous Systems
Prediction of Rolling Force Based on a Fusion of Extreme Learning
Machine and Self Learning Model of Rolling Force . . . . . . . . . . . . . . . . 3
AZiGuLi, Can Cui, Yonghong Xie, Shuang Ha, and Xiaochen Wang
User Behavior Profiles Establishment in Electric Power Industry . . . . . 12
Wei Song, Gang Liu, Di Luo, and Di Gao
The Optimal Pan for Baking Brownie . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Haiyan Li, Tangyu Wang, Xiaoyi Zhou, Guo Lei, Pengfei Yu,
Yaqun Huang, and Jun Wu
Intelligent Steering Control Based the Mathematical Motion Models
of Collision Avoidance for Fishing Vessel . . . . . . . . . . . . . . . . . . . . . . . . 27
Renqiang Wang, Yuelin Zhao, Keyin Miao, and Jianming Sun
Dynamical Analysis of Fractional-Order Hyper-chaotic System . . . . . . . 36
Junqing Feng and Guohong Liang
Evaluating the Performance of the Logistics Parks:
A State-of-the-Art Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Yingxiu Qi, Yan Sun, and Maoxiang Lang
A Tabu Search Algorithm for Loading Containers
on Double-Stack Cars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Zepeng Wang, Maoxiang Lang, Xuesong Zhou, Jay Przybyla,
and Yan Sun
Optimization Model Under Grouping Batch for Prefabricated
Components Production Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Chun Guang Chang and Yu Zhang
Air Quality Evaluation System Based on Stacked Auto-Encoder . . . . . . 63
Yuxuan Zhuang, Liang Chen, and Xiaojie Guo
ix
x Contents
E-Enabled Systems
Stock Market Forecasting Using S-System Model . . . . . . . . . . . . . . . . . 397
Wei Zhang and Bin Yang
Online Detection Approach to Auto Parts Internal Defect . . . . . . . . . . . 404
Lin Mingyu and Ju Jianping
Globalized Translation Talent Training Model based
on Artificial Intelligence and Big Data . . . . . . . . . . . . . . . . . . . . . . . . . . 410
Fang Yang
A Study on the Relationship between Enterprise Education
and Training and Operational Performance—A Cases Study
of a Multinational Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Yung Chang Wu, Lin Feng, and Shiann Ming Wu
Online Handwritten Character Recognition of New Tai Lue Based
on Online Random Forests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
Yong Yu, Pengfei Yu, Haiyan Li, and Hao Zhou
Web Information Transfer Between Android Client and Server . . . . . . 435
Baoqin Liu
Real-time Dynamic Data Analysis Model Based
on Wearable Smartband . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
Xiangyu Li, Chixiang Wang, Haodi Wang, and Junqi Guo
Investigation and Analysis on the Influencing Factors of Consumers’
Trust to Fresh Agricultural Products in E-commerce . . . . . . . . . . . . . . 450
Yipeng Li and Yong Zhang
A PROUD Methodology for TOGAF Business Architecture Modeling . . . 457
Feng Ni, Fang Dai, Michael J. Ryoba, Shaojian Qu, and Ying Ji
xiv Contents
Intelligent Systems
A Dynamic Node’s Trust Level Detection Scheme for Intelligent
Transportation System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
Ge-ning Zhang
An Application of Multi-Objective Genetic Algorithm Based
on Crossover Limited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 677
Shi Lianshuan and Chen YinMei
A Method for Extracting Objects in Physics Using Unit
Semantic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684
Yanli Wang and Pengpeng Jian
A Novel Matching Technique for Two-Sided Paper
Fragments Reassembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
Yi Wei, Lumeng Cao, Wen Yu, and Hao Wu
An Improved Memetic Algorithm with Novel Level Comparison
for Constrained Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 698
Xinghua Qu, Wei Zhao, Xiaoyi Feng, Liang Bai, and Bo Liu
FPGA Based Real-Time Processing Architecture for Recurrent
Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705
Yongbo Liao, Hongmei Li, and Zongbo Wang
Publication Recommendations of Manuscripts Using Improved
C4.5 Decision Tree Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 710
Didi Jia, Wu Xie, Zhiyong Chen, and Baohua Qiang
Contents xvii
Design Replica Consistency Maintenance Policy for the Oil and Gas
Pipeline Clouding SCADA Multiple Data Centers Storage System . . . . . 715
Miao Liu, Jia Shimin, and Yuan Mancang
SPSS-Based Research on Language Learning Strategy Use . . . . . . . . . . 722
Yang Xu
Study on Intelligent Monitoring of Power Transformer Based
on FPGA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 728
Fu-Sheng Li, Xin-Dong Li, Hong-Xue Bi, and Guang Jin
Mutual Fund Performance Analysis Using Nature Inspired
Optimization Techniques: A Critical Review . . . . . . . . . . . . . . . . . . . . . 734
Zeenat Afroz, Smruti Rekha Das, Debahuti Mishra,
and Srikanta Patnaik
Others
Synthesis Algorithm based on the Pre-evaluation of Quantum
Circuits for Linear Nearest Neighbor Architectures . . . . . . . . . . . . . . . . 749
Dejun Wang, Zhijin Guan, Yingying Tan, and YiZhen Wang
Study of an SIR Epidemic Disease Model with Special Nonlinear
Incidence and Computer Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 756
Xiuchao Song, Miaohua Liu, Hao Song, and Guohong Liang
An Improvement Response Surface Method Based on Weighted
Regression for Reliability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 761
Xingchen Yu and Zhangxue Gang
The Greedy Algorithm and Its Performance Guarantees
for Solving Maximization of Submodular Function . . . . . . . . . . . . . . . . 767
Yunxia Guo, Guohong Liang, and Jia Liu
Model of Evaluation on Quality of Graduates from Agricultural
University Based on Factor Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 772
Xuemei Zhao, Xiang Gao, Ke Meng, Xiaojing Zhou, Xiaoqiu Yu,
Jinhua Ye, Yan Xu, Hong Tian, Yufen Wei, and Xiaojuan Yu
Distributed Collaborative Control Model Based on Improved
Contract Net . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 779
Zhanjie Wang and Sumei Wang
K Distribution Clutter Modeling Based on Chebyshev Filter . . . . . . . . . 785
Bin Wang and Fengming Xin
Influence of Liquid Film Thickness on Dynamic Property
of Magnetic-Liquid Double Suspension Bearing . . . . . . . . . . . . . . . . . . . 792
Zhao Jianhua, Wang Qiang, Zhang Bin, and Chen Tao
xviii Contents
Abstract. Aiming at the rolling force model of hot strip rolling mill, the
forecasting technique based on ELM (extreme learning machine) is presented in
this paper. Initially, the variables associated with control rolling are identified by
the analysis of the traditional formula of rolling force, in order to ensure the
effectiveness of the model, and then apply ELM network to predict models.
Production data is applied to train and test the above network, and compare with
the modified calculated value of rolling force, which got from the self-learning
model of rolling force. The results show that the thickness can be predicted more
rapidly and exactly, which can meet the actual demand of production, when this
model and the rolling force learning model are integrated.
1 Introduction
Today, hot rolling production has more strict requirements on the dimension of steel
strip and the precision of plate-shape, in particular for preset of finishing block. And the
precision of finishing block preset be the center of the control model of finishing block,
rolling force involves a series of domain of operation, such as device parameter, rolling
parameter of adjustment and rolling enforcement [1]. And rolling force is related to the
thickness model control, strip shape model control and temperature model control,
therefore the control of rolling force is key to hot rolling production. The traditional
rolling force is counted according to the mathematical model, and the certain physical
quantities of mathematical model is counted according to the experiment many times,
which make the accuracy of forecast not high enough.
And ELM Algorithm have absolute advantages in such sectors as precision and
predicting speed. Therefore, paper will apply the advantage of ELM to rolling force
prediction. Rolling force in the every pass set different values according to the different
rolled piece parameters, because the multiple passes are set on rolling force. For
precision rolling, it need to set the rolling force of seven pass, so precision rolling need
to predict the rolling force of seven pass, in turn, according to ELM [2].
2 ELM Algorithm
ELM Algorithm is a simple and effective single invisible layer feed back network
SLFNs learning algorithm, which is firstly proposed by Prof. G.B. Huang [3] in 2004.
The learning rate of the tradition feed-forward [4] neural network far less than the
actual demand, which has been caused by two reasons. The reason one is that the
slowly learning algorithm, which based on gradient, has been widely applied to the
training of neural network, the all network parameters must be repeat adjusted for each
iteration. One the other hand, SVN and transformed smallest squared SVN has applied
to binary classification due to its powerful classification capability, and the traditional
SVN can not be used directly in regression and multi-class classification, which the
problem has not been solved though having proposed different SVN variant.
ELM turned to broad SLFNs, and hidden layers of ELM need not to adjust.
Parameters of hidden nodes is not only independent of training data, but also is
independent of each others [5]. The standard feed forward neural network having
hidden nodes has the ability to the separation and common approximate. ELM features
high efficiency learning, and generalization ability is better than traditional feed for-
ward neural network [4].
Taking the typical structure as an example, as shown in Fig. 1, the network is
comprised of three parts: input layer, hidden layer and output layer.
Figure 1, in the model, there are n input variable, it means that nerve cell number of
output layer is n. This indicates that the input vector of sample is x1 ; x2 ; . . .; xn , the
variable of output sample is m, this indicates that the output vector of sample is
y1 ; y2 ; . . .; yn . The connection of neighboring layers is determined by joining factor x
and b, and the linked weights matrix are formed by these factors [5].
Prediction of Rolling Force Based on a Fusion of ELM 5
Hidden layers nodes of n and standard SLFNs math expression of active function
g(xÞ is as followed:
XN 0
i¼1
bi gðxi xj þ bi Þ ¼ oj ; j ¼ 1; . . .; N ð2:1Þ
Hb ¼ T ð2:3Þ
H is output matrix of hidden layer [3], the number of hidden layer should be much
0
less than the number of sample in most cases, it means that when N N, H is
rectangle matrix [5]. When learning algorithm, H can be calculated and can be made
6 AZiGuLi et al.
stable through given the any value of x and b, and parameter b should be determined.
0
Through formula (2.3), the least-square solution b can be get:
0
Hb T ¼ minkHb T k ð2:6Þ
The minimum norm and the least-square solution can be get according to the
formula (2.3):
0
b ¼ H þT ð2:7Þ
A few important property can be get according to the generalized inverse matrix:
0
Least training error. The least-square solution of formula (2.3) is b from formula
(2.7). Although almost of learning algorithm hope to reach least training error, most of
it can not be done because of local minimum value and infinite iteration algorithm.
0
Weight minimum norm and optimal generation ability. b from formula (2.7) is
minimal norm of the least-square solution. For SLFNs, magnitude of weight is very
0
important, the smaller weight, the more powerful generation ability. It means b has
more better generation ability.
0
The minimal norm b of the least-square solution from (2.3) is unique.
The sample is trained by ELM [3], according to the random weight x and threshold
b, output matrix H can be get according to the set of hidden layers nodes and activation
function, corresponding prediction models can be get according to the formula of feed
forward network. Mainly ELM algorithm including:
Input weight x, b are given randomly, and corresponding hidden layers nodes are
get according to the real cases.
Select activation function g(X), calculate the hidden layer output matrix H [2].
0 0
Calculate the output weight b ,b ¼ H þ T.
For ELM, input weight x and threshold b can be generated at random, the nodes of
hidden layers are adjusted correspondingly to reach optimal output conclusion. This
study shows that compared with traditional BP network, ELM has the excellence of
faster learning speed, better generalization ability, and avoid the traditional algorithm
problems, including locally optimal tragedy, sensitive selection of learning rate and so
on [2].
So far, ELM has obtained approval by most researchers, and has been used in so
many fields, such as biology identification technology, image-processing, signal pro-
cessing, big data analysis, robot and internet. ELM could be used in the following
respects:
ELM can be trained in hidden layers feed forward network of widespread type, the
one layer of ELM [4] has been trained by hidden layer, which used in characteristics
learning, cluster, regression and classify. The entire network can be seen as no
adjusting ELM.
ELM can move into many local regions of multi hidden layer feed forward net-
work, and work together with other study architecture or model.
Prediction of Rolling Force Based on a Fusion of ELM 7
Each hidden layer nodes of ELM are sub-networks formed by some nodes, and
hidden layer can form local receptive domain [3].
In each hidden layer, the connection from the input layer to hidden layer node can
be full connection, and random local connection can be generated by the connection
according to the different continuous probability distribution. For the network [5], if a
limited number of hidden layer nodes and the connection have changed, the power of
the network still can make stable.
Compared with present mainstream president algorithm, the most advantage of
ELM is that the hidden neurons of the entire learning machine can not adjust, and have
high efficiency. Some algorithms of neural network have been applied to the steel
industry, like BP, RBF and Hopfield, but for ELM, there is a little application in steel
production industry for now. In this paper, ELM is applied to the steel rolling pro-
duction, and together with other methods, hot continuous rolling exit thickness and
rolling force are used to model for guiding steel production [5].
Rolling force can be calculated through some combination of the mathematical model
and empirical data, therefore, there are a lot of formula for rolling force. For now, the
correlative formula of rolling force has focus on the SIMS formula of OROWAN, and
researchers use the formula to calculate result in different rolling force condition, the
get different deformation formula, the original rolling force formula of SIMS follows
[5]:
0
P ¼ Blc Qp KKT ð3:1Þ
By some of these formulas and existed relationship among all variables, corre-
sponding physical makeup of the test samples are found by the database of Beihai steel
production. Because set-up calculation of Beihai steel production, modified calculation
and adaptive learning in steel production, that some of data need be discarded and the
measurement data need be chosen as samples, when samples are chosen. And
according to the corresponding formula of rolling force, variables are chosen such as
roll speed, roll rotate, forward slide coefficient, back slide coefficient, rolling force
measured value, FET measured temperature, FDT measured temperature, exit thickness
gauge measured value (deviation), exit width gauge, roll gap measured value,
8 AZiGuLi et al.
bend-roller force measured value, roll speed measured value, median width slab, entry
thickness measured value, entry temperature, exit temperature and roll gap.
The corresponding technology parameters for the rolling force model of finishing mill
is different depending on the site environment and the equipment situation [6]. In order
to ensure the shape and thick of the production, the computer of the model need be
optimized based on the data of the actual rolling. The rolling force method is modeled
according to the analysis of the deformation resistance and the coefficient of stress
state. And the initial parameters are set, AGC control is used for model, which can
achieve continuous self learning, to adjust rolling force. As compared with self learning
of the other parameters, the biggest difference is that rolling force can not compare the
measured rolling force of the N pass steel with the original estimate rolling force.
Because bounce is a remarkable feature of mil, when measured rolling force become
the ups and downs at the change, because of the elastic deformation, the value of actual
thickness is more or less than the value of the original prediction thickness. So the
value set in rolling force and the measured value of rolling force can be get in two
different technology parameters, which can not be compared. And the self learning
model used the measured rolling force to correct the model parameters, rolling force
model recalculated the rolling force depending on correction factor, this step calcula-
tion is called correction calculation. So compared with the measured value and the
value of rolling force got from the correction calculation, the feature of the model can
be reflected.
Through the above process of the rolling force model, in production, initially, when
slab is required to roll through the first finishing mill, the measured value of rolling
force in the first pass can be get, and the parameters of rolling force model are corrected
by the model according to the measured value of rolling force in the first pass, then, the
value of rolling force is recalculated by the original model, and the parameters of the
production line in the second pass are set by the value of rolling force which was got
according to the correction calculation [8], the second roll is finished. When the second
roll is finished, the rolling force are corrected to calculate by repeating the step of the
first roll, finally, the value of the rolling force is controlled by repeating the self
learning in order to the guide the steel production.
According to the above the finishing mill principle, initially, through the formula of the
rolling force, affecting factors about the rolling force are found and regarded as the
input variable in order to predict for the rolling force in the first pass, the measured
value of rolling force in the first pass which is got in the steel production and the
corresponding affecting data about the rolling force are regarded as the input variable
into the next ELM model in order to predict the rolling force in the second pass, by
Prediction of Rolling Force Based on a Fusion of ELM 9
analogy, when predicting the rolling force in the seventh pass, the prediction rolling
force in the top six pass is got by the model in order to guide the production, so the
measured value of rolling force in the top six pass which is got in the steel production
and the corresponding affecting data about the rolling force are regarded as the input
variable into the next ELM [6] model in order to predict the rolling force in the seventh
pass, finally, the all of the prediction data in the seven passes are got, the multiple ELM
models superposition are realized.
First of all, divide the 522 filtrated groups of data which is related to the rolling
force into training set and test set, with 472 group as the training set, the remaining 50
group as a test set, to predict the first pass rolling force of the finishing mill. Due to the
number of nodes in hidden layer can produce certain influence on prediction accuracy,
so many experiments are needed to determine the optimal number of nodes. The
conclusion can get through the experiment, when the number of hidden layer nodes is
80, network performance achieves the best. The activation function is “sig” function.
Based on the data, the study found some dimensions of the data which is related to the
rolling force are small, while some are very large. If different dimensions are not
processed, the rate of convergence of the network will be reduced. First, process the
input data as normalization, put the rolling force data to be quantitative without the
outline, and then carries on the training and obtain the comparison between the actual
rolling force with predict rolling force model.
From the picture the most of points of first time prediction of rolling force and
actual rolling force are close, occasionally there are some larger errors. The reason of
these errors is the rolling force value is bigger, so the influence of the amplitude is
bigger. For the traditional rolling force model, the comparison with the measured
rolling force can not be done by setting calculation of rolling force. Correct calculation
of rolling force and the measured rolling force should be compared [7]. Therefore
corrected average standard deviation of the calculated rolling force and ELM prediction
model can be seen in the Fig. 2.
Fig. 2. The predicted value of the rolling force and the actual value of the rolling force for ELM
10 AZiGuLi et al.
The first mill prediction precision of rolling force is slightly higher than the pre-
cision of rolling force obtained by modified computing. Because correction calculation
of rolling force is for the guidelines to rolling mill, so the predicted rolling force can be
used in place of the correct calculation of rolling force to guide production. Because the
order of magnitude of rolling force unit is big, the experiment will increase the
numerical calculation, so orders of magnitude unit need to be downsized, and then
according to the calculation results restore the order of magnitude of the corresponding
unit. Make the first mill rolling force production actual value from ELM model with
other variables as the input of the second mill rolling force prediction model [7], the
actual value of rolling force by second mill again make the third mill modeling and so
on to get seven mills of ELM superposition model. The predicted results compare with
the measured values and get the average standard deviations, and compare with average
standard deviation fixed calculated (Table 1).
The first mill prediction precision of rolling force is slightly higher than the pre-
cision of rolling force obtained by modified computing [8]. Because correction cal-
culation of rolling force is for the guidelines to rolling mill, so the predicted rolling
force can be used in place of the correct calculation of rolling force to guide production
[8]. Because the order of magnitude of rolling force unit is big, the experiment will
increase the numerical calculation, so orders of magnitude unit need to be downsized,
and then according to the calculation results restore the order of magnitude of the
corresponding unit. Make the first mill rolling force production actual value from ELM
model with other variables as the input of the second mill rolling force prediction
model [5], the actual value of rolling force by second mill again make the third mill
modeling and so on to get seven mills of ELM superposition model [4]. The predicted
results compare with the measured values and get the average standard deviations, and
compare with average standard deviation fixed calculated (Table 2).
Table 2. The error value of each pass rolling force for precision rolling
Pass The prediction of The correction
rolling force calculation of rolling
force
Time (s) RMSE (103) Time (s) RMSE (103)
1 0.004 0.03708315 – 0.08370321
2 0.001 0.03686208 – 0.05952837
3 0.001 0.05461632 – 0.04789749
4 0.001 0.03943099 – 0.02624345
5 0.001 0.04748756 – 0.03547987
6 0.001 0.03010246 – 0.03407099
7 0.001 0.02245528 – 0.07047431
Prediction of Rolling Force Based on a Fusion of ELM 11
6 Conclusion
From the above experimental data, the use of ELM for continuous prediction of seven
passes in the results, the first and second mills of the rolling force prediction accuracy
compared with the correction precision of the calculation model is higher; the third to
the fifth mills of prediction precision deviation is bigger [8], the deflection prediction
accuracy is lower than those of correction calculation model, the effect is poorer. Mills
of 6 and 7 of the predicted results compared with the correction results found that 6 and
7 of the finishing passes by using machine learning to predict the effect of is better than
the effect of the modified calculation.
If all seven passes of finishing mill group through ELM rolling force prediction
model, then in the third to the fifth mills due to big errors may lead the final exported
steel does not meet the actual requirements. So according to the above experimental
results to improve the model, through the ELM and self learning of system integration
method to predict the rolling force, build the model as shown in Fig. 3. Two mills
before can use ELM overlay to predict the rolling force, and according to the predicted
rolling force to guide production [9]. The third to the fifth mills on-line correction can
through the system of self learning, Then the sixth, seven mills again apply ELM
rolling force prediction model.
Through improving model to forecast the rolling force, the time of forecast can be
greatly shortened and solve the real-time demanding of rolling technology problem,
and get the strong robustness. High accuracy model will has a broad application
prospect.
References
1. Sun, Y.K.: Model of Control Hot Strip Mill, pp. 124–163. Metallurgical Industry Press,
Beijng (2002)
2. Huang, G.B., Zhu, Q.Y., Siew, C.K.: Extreme learning machine: theory and applications.
Neuro-Computing 70, 489–501 (2006)
3. Huang, G.B., Zhu, Q.Y., Siew, C.K.: Extreme learning machine: a new learning scheme of
feed-forward neural networks. In: Proceedings of International Joint Conference on Neural
Networks (IJCNN 2004), pp. 985–990 (2004)
4. Lonen, J., Kamarainen, J.-K., Lampinen, J.: Differential evolution training algorithm for feed-
forward neural networks. Neural Process. Lett. 7(1), 93–105 (2003)
5. Huang, G.B., Chen, L., Siew, C.K.: Universal approximation using incremental constructive
feed-forward networks with random hidden nodes. IEEE Trans. Neural Netw. 17(4), 879–892
(2006)
6. Huang, G.B., Chen, L.: Convex incremental extreme learning machine. Neurocomputing 70,
3056–3062 (2007)
7. Xie, J., Jiang, S., Xie, W., et al.: An efficient global K-means clustering algorithm. J. Comput.
6(2), 271–279 (2011)
8. Al-Zoubi, M.B., Hudaib, A., Huneiti, A., et al.: New efficient strategy to accelerate k-means
clustering algorithm. Am. J. Appl. Sci. 5(9), 1247–1250 (2008)
9. Gaemperle, R., Mueller, S.D., Koumoutsakos, P.: A parameter study for differential evolution.
In: Grmela, A., Mastorakis, N.E. (eds.) Advances in Intelligent Systems, Fuzzy Systems,
Evolutionary Computation, pp. 293–298. WSEAS Press (2002)
User Behavior Profiles Establishment
in Electric Power Industry
Abstract. The big data of electric power industry contains the information of
users’ values, credits, and behavior preferences. On the basis of the data, electric
power companies can provide personal services as well as increasing profits. In
this paper, we proposed a labeling system based on the clustering algorithms and
Gradient Boost Decision Tree (GBDT) algorithm to establish user behavior
profiles for State Grid Group of China, including basic information labels,
behavior labels, behavior description labels, behavior prediction, and user
classification. The experimental results showed that the approach can describe
the behavior features of users in the electric power industry effectively.
1 Introduction
There are many kinds of behaviors in electric power industry including electricity
consuming, paying, capacity changing, illegally electricity using, etc. Although the
electric power companies have a lot of users’ behavior data, they are numerical and
hard to understand directly. Some rules are needed to turn the initial numerical data to
semantic labels which could easily illustrate the users’ feature. Here are some of the
labels: ‘Zhangjiakou’, ‘Low monthly consumption’, ‘capacity change frequently’,
‘Recover from capacity suspend every 7 months’, ‘High risk of arrearage’, ‘Low user
value’. These labels are called user profiles. Profiling the users is the basis of providing
the personal services, and its applications in the electric power industry are various.
First, companies can select users by labels, which are more detailed and semantic. Also,
it’s easy to recognize the occurrence of the user behaviors so that companies can easily
monitor the user behaviors and give advise. At the same time, as the companies handle
the feature of the user behaviors, precision marketing will be reasonable and the
companies can arrange business better. What the most important is, according to the
user profiles of electric power industry, the companies can get the semantic credit
report of the users. User profile is commonly used in IT industry, but we can not build
the user profiles in the electricity industry using the same approach in the IT industry
Language: French
PAR
GASTON LEMAY
PARIS
G. CHARPENTIER, LIBRAIRE-ÉDITEUR
13, RUE DE GRENELLE-SAINT-GERMAIN, 13
1879
Tous droits réservés.
PARIS. — IMPRIMERIE Vve P. LAROUSSE ET Cie
19, RUE MONTPARNASSE, 19
A mes compagnons de voyage
A l’État-major
et à l’Équipage de la JUNON
Gaston Lemay.
A MONSIEUR GEORGES BIARD
LIEUTENANT DE VAISSEAU
G. L.
DE MARSEILLE A GIBRALTAR
En mer, 2 août.
3 août.
4 août.