Mei C., Zhou J., Peng X. Simulation and Optimization of Furnaces and Kilns for Nonferrous Metallurgical Engineering
Подождите немного. Документ загружается.
10 Multi-objective Systematic Optimization of FKNME
the users’ input information, harmonize communication in each component, and
outputs the decision results and relates information according to users’
requirements.
Data component is made up of a database and a database management system,
and all needed data of decision-making are lodged in the database; the database
management system is used to edit, amend and organize data, and connect
database with other components. Data component supplies the data service for a
decision making.
Fig. 10.4 Structure of decision support system
Model component is made up of a model base and a model base management
system. Modal base is used to lodge the models for a system decision; the model
base management system connects knowledge reasoning components and the
model base, and manages model bases by establishing, adding and deleting,
amending and calling models in it.
Knowledge reasoning component is composed of the knowledge base, a
knowledge base management subsystem, reasoning machines. It is the core of a
decision support system. Knowledge base stores knowledge needed in forms of
production rule, predicate logic, semantic network, frameworks, units and script
and so on. As the main control unit of system, knowledge base management
subsystem controls the synthetical application of all kinds of data, knowledge,
model and method, and unifies all components. Reasoning machine can
accomplish searching and matching for knowledge.
10.4.2.2
Example: intelligent decision support system for nickel smelting process
The technological process of some smelter is “submerged arc furnace-converter-slag
cleaning furnace”
(Xiao and Xie, 1997), as shown in Fig. 10.5.
Obviously, it is a typical complex system with multi-variables, nonlinearity and
big random interference. In order to realize energy saving and consumption
reducing of production process, in this paper, an intelligent decision support
system of nickel smelting process is developed based on the optimal decision
Xiaoqi Peng, Yanpo Song, Zhuo Chen and Junfeng Yao
making model for a smelting process (Peng et al., 1994) introduced in Sections 4.3
and 4.4 and endpoint prediction model for the converting process (Peng, 1998),
according to the technical route of “mathematical model
ühologram simulationü
synthetical optimization”. It provides decision support for various posts and
various processes in full duration time from many aspects such as theory,
calculation, scheduling, operation experience and so on. Practical applications
showed that this system could guide production operation effectively, improve
production performance indexes, obviously decline the unit production energy
consumption, reduce the valuable metal loss in slag and auxiliary raw material
consumption, promote metals resource utilization ratio. As a result, significant
economic and social benefit were obtained. Besides, the system could also offer
lots of statistical analysis charts on smelting processes, which provided decision
makers with strong assistant tools for knowing production performance, analyzing
influence factors of production indexes, and taking effective measures to improve
production conditions (Peng and Mei, 1996; Mei et al., 1994; Peng et al., 1994,
1995).
Fig. 10.5 Technological process in smelting workshop
Comparing to the production indexes before the application of the system, the
improvements achieved are as follows:
a) The electric energy consumption per ton of calcine of submerged arc furnace
was reduced by 14 kW
gh, 2.3% lower than before; the nickel in smelting slag
reduced 0.014, 6.4% lower than before.
b) The weight of furnace charge smelt in converter was increased by 111%.
c) The electric energy consumption per ton of converter slag was reduced by
182kW
gh, 39.6% lower than before; the cobalt in smelting slag of electric furnace
for cleaning slag was reduced by 0.022%, 23.7% lower than before; the nickel in
smelting slag of electric furnace for cleaning slag was reduced by 0.043%, 61.4%
lower than before.
The results mentioned above showed that energy saving and consumption
10 Multi-objective Systematic Optimization of FKNME
reducing effects of the system were obvious, and it reached the aim of the
design.
Besides, because of the application of the system, the furnace conditions were
stabilized and its lifetime was prolonged, the accident rate was reduced, the
consumption of raw materials such as quartz, electrode paste and refractories were
reduced also.
The overall structure of the intelligent decision support system for nickel
smelting process is shown in Fig. 10.6.
Fig. 10.6 Intelligent decision support system for nickel smelting process
Man-computer interface communicates with users in session manner by
multilevel menu driving.
In database, all the data for production decisions and technological
managements are stored.
In knowledge base, the decision knowledge in forms of production rules, fuzzy
decision rules and optimal samples is stored, which can be modified automatically
by the system.
In model base, the decision making models are stored. Model base management
subsystem has self-learning and self-adaptive ability.
Method base and its management system constitute the method base subsystem.
All kinds of algorithms needed by the system operation are stored in the method
base; method base management system links the knowledge base subsystem and
the method base, and controls addition, deletion, modification and calling of
methods in method base.
Technology management subsystem
analyzes technological parameters and
manages the incoming material testing files by using the data in database, and
accomplishes file management for process equipments, process technologies,
Xiaoqi Peng, Yanpo Song, Zhuo Chen and Junfeng Yao
process accidents, process repairing, quality management and technology reforms.
The structure of the model base subsystem is shown in Fig.10.7:
Fig. 10.7 Model subsystem
a) The material composition forecasting models are used to forecast the
composition of materials in production, which overcomes the time-lag of the
composition test data and supplies the necessary composition data for a optimal
decision making. According to the prediction error and the recent composition
change, the system can modify the forecasting model automatically and makes the
forecasting results credible and accurate.
b) The fuzzy adaptive optimization decision model of the submerged arc furnace
production is used to optimize the input materials and the operation parameters for
ore-smelting furnace. Because the modeling data come from a knowledge base,
while the knowledge base is self-learning and self-adaptive, the model is a fuzzy
adaptive optimal decision making modal.
c) The fuzzy neural network optimization decision model of the slag cleaning electric
furnace production is used to optimize the input materials and operation parameters of a
cleaning furnace. The model is self-learning and self-adaptive because the final learning
results are always the last decision-making model in system modeling.
d) The converting endpoint prediction and control model (Peng, 1998) is used to
predict converting endpoint and optimize converting parameters.
e) The production materials balance and heat balance calculation model
10 Multi-objective Systematic Optimization of FKNME
calculates the materials balance and the heat balance of furnaces, which can
simplify users’ analyzing the flowing direction of materials and heat, and take the
measures to improve production conditions and technology.
The structure of a method base subsystem is as Fig. 10.8, the related algorithms
that system running needs are stored in the method base:
Fig. 10.8
Method base subsystem
a) Fuzzy decision model recognition algorithm. It is used to establish the
optimization decision model of proportioning and operation parameters in
submerged arc furnace production.
b) Fuzzy neural network weights learning algorithm. It is used to establish the
optimization decision model of proportioning and operation parameters in slag
cleaning electric furnace production.
c) Time-series filtering prediction algorithm. It is used to establish the materials
composition prediction model and the converting endpoint prediction and control model.
d) Algorithms for solving linear equations. They are used to establish the
material balance and heat balance calculation model of furnaces.
e) Regression analysis algorithms, including generally used regression analysis
algorithms such as multivariate linear regression, power series regression,
exponent regression and logarithm regression. They are applied in the regression
analysis and related analysis for all kinds of production data, which makes the
users understand the production more easily and quickly.
The structure of technology management subsystem is as Fig.10.9, the main
Xiaoqi Peng, Yanpo Song, Zhuo Chen and Junfeng Yao
functions are as following:
Fig. 10.9 Technology management subsystem
a) Parameters control analysis. It provides a mass of statistic and analysis figures
and tables about nickel smelting production of the whole smelting workshop, and
gives all sorts of composition fluctuation figures of materials (calcine, quartz,
vulcanizing agent, etc.), the relation curve of valuable metals content in slag (content
of cobalt in slag, content of nickel in slag) and other content of slag compositions,
and the power utilization curve of the electric furnace in the given time range, which
helps the users know the dynamic production process directly and visually, and
improve the product condition and technology in time.
b) Balance calculation. It accomplishes the material balance and heat balance
calculation of a nickel smelting submerged arc electric furnace, converter and slag
cleaning electric furnace productions, and can print the material balance table, heat
balance table, the chemical elements balance table and the composition tables of all
sorts of materials for product plan and decision makers’ reference.
c) Archives of incoming material inspection. It manages the information such as
incoming data, product batch, quality, composition of material from other
workshops by a computer.
d) Archives of quality management. It manages all kinds of quality management
information such as members in a quality management team of each sections and
process, their work performance and production quality change by a computer.
e) Archives of technology equipment. It manages information of technology
equipment in workshop by a computer.
f) Archives of technology. It manages information of all technologies in workshop
and information of new technology and new equipment in the same domestic
industry by computer.
g) Archives of technology improvement. It manages technology improvement
10 Multi-objective Systematic Optimization of FKNME
information of all sorts of equipments and technologies in workshop by a computer.
h) Archives of technology accident. It manages all technology accident information
in workshop by computer.
i) Archives of technology check. It manages all technology check information in
workshop by computer.
Therefore, the main functions of the intelligent decision support system are as
following:
a) It collects formatively production data, statistically analyzes them and prints
production management and statistic reports.
b) Basic technical analysis of heat engineering process: calculate the elements
distribution, material balance, heat distribution and heat balance in the smelting
process, provide the material balance table, heat balance table, chemical element
balance table and phase composition table of all sorts of materials so as to analyze
flow direction of material and heat in smelting and improve the production
condition and technology.
c) Dynamic analysis of production index: perform dynamic analysis, regression
analysis and related analysis to the main technology-economy indexes of
individual furnaces, draw production index fluctuation figures, control figures,
distributing figures and histograms so as to make decision makers realize
production condition in time, analyze the factor which impacts the production
indexes and improve the production condition and technology.
d) Man-machine interactive optimizing decision-making in production with aim to
save energy and reduce consumption: determine the optimized material proportion,
production power of electric furnace and output amount of low-nickel matte and slag
according to the submerged arc furnace production fuzzy self-adaptive optimization
decision-making model; predict converting endpoint according to product condition
of low-nickel and provide optimized converting strategy; determine the optimized
material proportion, the load current and its duration in the smelting stage and
cleaning stage, and output amount of cobalt matte and cleaned slag according to slag
cleaning furnace fuzzy neural network adaptive optimal decision making model.
The whole decision-making process is man-machine interactive, users can
discretionarily change any decision-making result according to actual production
condition, and system will make new optimal decision stages for reference
automatically according to users’ requirements.
10.4.3
Online optimization system
Because of the changes in the environment, the aging of equipment or catalyst, the
fluctuation of raw material composition and so on, it is impossible for offline
optimization to keep industrial process running optimally all the time. Hence,
Xiaoqi Peng, Yanpo Song, Zhuo Chen and Junfeng Yao
online optimization system is needed.
10.4.3.1
Optimal control problem
Optimal control problem can be described as follows:
Given a controlled system as:
(, , )ft
=
&
x
xu (10.43)
Initial state of the system at time t
0
is:
x (t
0
) = x
0
(10.44)
where,
x is the state vector of the system;
x
&
is the derivative of the state vector; u
is the control vector; t is the time; the system usually is nonlinear and time-varying.
Given the final state or goal set at the end time t
1
is:
x(t
1
)=x
1
or g(t
1
,x(t
1
))=0 (10.45)
Therefore, there needs searching an admissible control Ut
∈)(u , t
0
İtİt
1
, where
U is admissible control set, driven by an admissible control, by which the
controlled system starts from initial state
x
0
at time t
0
and reaches to the target
state
x
1
or target set at time t
1
; meanwhile, the performance evaluation function
(performance index) J should be minimum. This kind of problem is the optimal
control problem.
1
11
0
(, ()) (, (),())d
J
St t Lt t t t=+
∫
xxu (10.46)
The performance evaluation function J reflects the quality and performance of the
control system, and the first term S reflects the allowable deviation from the target,
called the terminal-value performance index; the second term reflects the dynamic
performance of the control system, called the integral performance index; the
control function
)(* tu that makes J get the minimum value is called the optimal
control, the roots track
)(* tx of the corresponding system state equation is
called the optimal track.
Therefore, the so-called optimal control is in fact the best control according to a
performance evaluation criterion, the process of solving optimal control problems
is the process of choosing an optimal control vector(or control project) from all
optimal control vectors (or control projects). The performance index confirms the
solving of the optimal control problem.
10.4.3.2
Introduction for common optimal algorithms
Researchers in math field and control field have carried out lots of studies on the
solutions of the optimal control problems, and have obtained great achievements.
Many solution methods have been used broadly in practice.
For a controlled system with an accurate mathematic model, the main math
methods to optimize it are classical differential method and variational method. In
1953~1957, when studying the multistage decision process optimization problem,
Bellman found that the optimization decision sequence in the multistage decision
10 Multi-objective Systematic Optimization of FKNME
process has properties as follows: no matter what the initial stage, the initial state
and the initial decision are, taking the stage and state formed by the first decision
as the initial condition, the subsequent decisions must constitute optimal decision
sequence for the subsequent problems, which is the famous Bellman Optimal
Principle; based on this, Bellman proposed dynamic programming solving for the
optimal control problems. In 1956~1958, enlighten by Hamilton Principle of
mechanics, Pontryagin proposed maximum principle on the foundation of classic
variational method. Dynamic programming and maximum principle play
important roles in classic optimal control theory (Peng and Mei, 1996).
There are different solutions for different types of optimal control problems, the
commonly used classic optimal algorithms include: variational method,
Pontryagin maximum principle, gradient method, conjugate gradient method,
Bellman dynamic programming, least square method, simplex method and so on
(Gao, 1991; Xiao and Xie, 1997; Peng et al., 1994).
Classic optimal control algorithms are mainly suitable for the deterministic control
system with an accurate mathematic model, while the structures and parameters of
most actual controlled objects are uncertain in some extent, and the change of a state
possesses some randomness because of the effect of a disturbance, so it is difficult to
solve this actual problems using the classic optimal control algorithms directly,
therefore, the idea of developing an adaptive control system is proposed.
So-called an adaptive control system is a control system which can measure
continuously or periodically dynamic performance of an object and performance
index of a system in the control process, then compare them with anticipant
dynamic performance and given performance index, and make decision
according to comparison result, that is, by changing structure and adjustable
parameters or producing a control signal, make the system run in an optimal or
suboptimal state no matter how the environment changes. The goal of the
adaptive control system is to eliminate the effects of structure and parameters’
disturbance on performance of the system. The structure of an adaptive control
system is as shown in Fig. 10.10.
Fig. 10.10 Structure of adaptive control system
Xiaoqi Peng, Yanpo Song, Zhuo Chen and Junfeng Yao
Since 1970s, model reference adaptive system (Landau, 1985), self-tuning control
system (Feng, 1986), learning control system (Chang, 1982; Katsuhiko, 1980) and other
adaptive systems have gotten significant development both in theory and in practice.
For complex controlled process of high orders, nonlinearity, long time delay,
time varying and strong random disturbance, it is difficult even impossible to
apply classic optimal control theory based on the accurate mathematic model.
Therefore, based on the study of the human brain structure and intelligence
characteristics, the idea of optimizing system by means of human-simulated
intelligent control is proposed, guided by this idea, lots of fruitful research and
develop work are performed (Peng et al., 1994; Zhao, 1996).
By simulating the qualitative reasoning, fuzzy reasoning and knowledge-learning
process of human brain, the theories and technologies of expert system, self-organizing
and self-adaptive fuzzy control (Li, 1993), neural network adaptive control
(Hu, 1993)
were proposed, which were applied to solving the optimal control problems of
complex process and achieved good performance.
10.4.4
Integrated system for monitoring, control and management
Because of complexity, mutagenicity and uncertainty of industrial production
processes, it is very difficult to implement overall automatic control. Applying
synthetically modern management and decision technology, information science
and technology, automation science and technology, system engineering
technology and artificial intelligence technology, integrating expert group, data
and information with computer technology, the computer information integrated
system for monitoring, control and management can be constructed. The system
integrates the information of three elements (that is, personnel, technology and
management), information flow and material flow, can be used to elaborate overall
advantages of enterprise and realize cooperative work, which has been an effective
approach to develop the current potential, improve production technology indexes
and improve the competitiveness for an enterprise.
10.4.4.1
System Structure and Working Principle
An integrated system for monitoring, control and management mainly includes:
object layer, measuring and control layer, supervision and scheduling layer,
information management and decision making layer and corresponding
communication network. Its structure is shown in Fig. 10.11.
Controlled object is the industrial production process. Various information
reflecting production state is sent to the measurement and control layer by certain
sensors and measurement instruments, production process are controlled directly
by measurement and control layer.